case study on demand analysis

• Browse All Articles
• Newsletter Sign-Up

DemandandConsumers →

No results found in working knowledge.

• Were any results found in one of the other content buckets on the left?
• Try removing some search filters.
• Use different search filters.

A step-by-step guide to product demand analysis in 2023

2020 showed us: even the steady sales of toilet paper can’t be taken for granted. Sudden surges or drops in demand happen across all product categories, from soap to software.

What is a product demand analysis?

Why conduct a product demand analysis, what are the types of product demand, how to conduct a product demand analysis in 5 steps, product demand analysis examples, product demand analysis best practices, picking the right market research tools.

But if you become an expert at product demand analysis, those highs and lows don’t catch you by surprise as often. In this guide, you’ll learn the foundations of product demand analysis for the marketplace of tomorrow.

With a product demand analysis, you try to get an accurate estimate of future sales of your product. It’s a way of understanding how competition, seasons and other relevant events affect the sales of a certain product. 

Product demand analysis can be done at various times – even for products that aren’t for sale yet. It’s not only based on past sales, demand can also be predicted based on changes in society, technological advancements and environmental changes. 

So yes, there are a lot of factors weighing in, and nobody can predict the future down to the last chocolate bar being sold. But gauging product demand is crucial for building a future-proof business. Here’s why.

The goals of your product demand analysis depend very much on at what stage your business or product is in. 

Validating ideas and financial planning

You could be doing exploratory market research and trying to find out if there’s a big enough market for you to enter with your product. And if there is, could you enter at a price point high enough to make your idea worth pursuing? 

Buying materials from suppliers

Product demand analysis is also important for businesses who heavily rely on secondary manufacturers or resources from external sources. Will you be able to get the necessary materials in time? 

Save money and work more efficiently

Saving money could also be one of the goals of performing a product demand analysis. Knowing when your product will be popular will help you better allocate your budget and manpower. Knowing when demand will be higher will help you plan the budget and timing for marketing campaigns, but also make sure you have enough employees on the work floor to handle the extra orders. 

Track product demand with our audience of 125 million

Attest makes any product demand analysis easy. Unlock new sources of growth and find out what product features your key customers want.

Product demand isn’t as straightforward as you’d think. Products and categories are connected to each other, and sometimes demand comes from where you were least expecting it. Here are the most important types of product demand with examples. 

• Direct demand: the simplest form of product demand is the demand for a final product. For instance, how many people are planning to buy a new smart TV.
• Indirect Demand is the demand for a product that is used to produce another product. Are you still following? A simple example is the rising demand for, let’s say, standing desks. The wood of the table would be the product that is in higher indirect demand, due to the popularity of the end product.
• Joint demand occurs when two products have a direct and positive correlation in demand. If people start buying more paint, they will start buying more canvases.
• Composite demand is the demand for products that can be used in more ways than one. For materials like the wood earlier, it’s important to know that it’s not just in demand for stand-up desks. It could also be used for millions of other products.
• Latent demand is the demand for a product that consumers can’t satisfy. There are three scenarios in which this happens:

1. The consumer doesn’t have the means to buy the product. They want it, but they can’t afford it. 

2. The product that consumers want isn’t available.

3. The consumer doesn’t know the product exists or doesn’t know a certain product fulfils their needs. 

When conducting your product demand analysis, it’s important to distinguish between these types of demand, but also to see the connections where possible. You might be able to benefit from composite or joint demand, while never having looked into that option.

Product demand analysis shouldn’t be wild guessing: it’s building a solid foundation of knowledge and structuring your research on top of that- and it’s a fundamental part of market research for new product development . Here are the steps:

• Define your market
• Assess the maturity of the market business cycle

Identify your market niche

Calculate market growth potential.

• Evaluate the competition 

Define your market 

Who could you be selling to, how much do those people have to spend, and who are they currently giving their money to? Those are crucial questions to ask when defining your market. 

Defining your market shouldn’t just be done in words, by explaining personas or target audiences – the numbers are crucial to this. 

Depending on the goal of your product demand analysis, you could also look at secondary markets for this part of your research. There might be some ground you could cover in markets where your product would be used by people who are not in your primary target audience. 

Assess the maturity of the market and business cycle

Some products seem to have an endless life cycle. The oldest beer brewery has been around since 1040, and we’re still drinking it – even though many other alternatives have entered the market. But there are also breweries that have gone bankrupt. What’s up with that?

The market and business life cycle are two important things to analyse. First of all, you want to get an idea of how steady the market is. Will you be entering a market that is in the growth phase, is it mature and stable, or is it heading towards decline?

The same should be done for your business: how mature is it? It’s important to make this a part of your product demand analysis. If your market is growing but you can’t keep up, how will you deliver enough products to actually be profitable?

Your market niche is the sweet spot of where there’s potential to enter and where you would be able to deliver. Not only in terms of physical products or practical services, it should also be a match when it comes to values and USPs. Consumers aren’t just looking for products, they are looking for brands they connect with. 

When doing a product market analysis, it’s important that you are specific and don’t paint a better picture by looking at the entire market, or too large a part of it. Eventually, you will have to choose a targeted message that will not speak to everyone in the market: keep this in mind when defining the real size of your market.

Find your niche fast with Attest

Our consumer segmentation filters and built-in audience of over 125 million people make finding your niche easy.

Now you have the data on the current situation: time to look into the future. To do that, we’ll start by going to the past. 

How has the market share been divided among your competitors ever since the market came to exist? What events and products affected these changes in the division? What was happening in related or similar markets in the same time period? 

Based on this information you can try to pinpoint patterns and find opportunities for growth in the future. Also keep societal factors in mind, like wages and costs of other possibly related products, and changes in tax.

Evaluate the competition

Chances are, your competitors are also performing some kind of product demand analysis while you are too. Maybe they also have a new product in mind that they want to launch, or they’re trying to increase their market share another way. 

Look at how they went about their previous product launches and how this was received by your target market. How were the sales numbers? What could you learn from them?

You can keep an eye on what’s being said by and about your competitors using smart competitor tracking tools . 

Is it magic? Is it witchcraft? Is it espionage? No, it’s artificial intelligence. 

One of the kings in product demand analysis is undoubtedly Amazon. How do they have that tiny screw in the right colour available for next day delivery, when all hardware stores in your area are out of stock?

A lot of it is thanks to artificial intelligence, and an exquisite product demand analysis. Amazon has mastered the craft of balancing human intelligence with human tasks. Product demand analysis on the scale they are doing would simply take too long to do manually, so where possible they let AI do the heavy lifting. In the background, their team is figuring out the details that make their supply chain so impressive. 

The team knows not only what products to have in stock, but also where. In 2013, they got a patent for ‘anticipatory shipping’. This technique helped them to get a product to the closest warehouse to you, even before you actually hit ‘buy now’.

Other parts of their product demand analysis are more common sense. Certain local products will only get stocked in relevant regions, and they look beyond the obvious seasonal changes. Sunscreen is important in summer, yes, but also in winter in places where, for example, families go on skiing holidays. 

How do you actually gather all that information we mentioned above? Let’s look at some tools and best practices that will help you get the most relevant and actionable insights regarding product demand predictions. 

Product demand surveys

To accurately estimate the demand for your product through a survey, it’s crucial that you ask the right questions. Think about how precise you would want people to answer, what data you need. Is it just numbers, or also about days or months? Will you ask them about alternative buying reasons for your product? 

Set a clear goal for your product demand survey and build the questions based on that, so you won’t miss a single piece of information. 

Another crucial element is defining quality respondents. Your product demand analysis survey should only be sent to people who match the criteria of actual buyers. This way you prevent getting data on latent demand, from people who would buy, but don’t actually have the money. 

Ask the right questions with the right survey

With templates for market analysis plus on-demand support from our research experts, we’re here to help you gather continuous product demand data, fast.

Experiments with price and offer

Some markets are unpredictable, mostly because they are new. If after your initial product demand analysis you find that the numbers are a tiny bit lower than you’d hoped for, try analysing what happens if you tweak your offer.

You can experiment with different price points, package deals or bundles, or other types of promotions. You can also try finding partnerships or products that go well with yours to create more demand, even if it’s just secondary. This could be especially helpful if you are just entering a new market, and you want to ride on the success of another product.

Consumer trends

To identify consumer trends, you can use tools such as Google Trends, look into frequent Amazon searches and keep an eye on trending topics on social media. 

With social listening tools, you can keep track of what people are saying about a certain product or brand, which will help you get data on what’s trending. Joining online communities or Facebook groups that are relevant to your market is also a great way to stay up to date on what’s happening. 

Don’t forget to also look into popular topics in specific regions if that’s relevant for you. When looking at specific keywords, you can get incredibly precise data that will drive your decision-making process forward. 

Still feel like product demand analysis is just a guessing game? With the right market research tools , you’ll see that getting accurate data is more a science than an art. Here are four of our favourite tools for market research that could help you getting your product demand analysis right:

• Attest : we had to! With our quick, easy-to-use market research platform you can find those hyper-relevant consumers and ask them those burning questions about their buying behaviour.
• Social Mention : to keep an eye on consumer trends, we highly recommend using social mention. This is a social listening tool that will give you first-hand data on what’s being said about your market across different platforms, without you having to monitor them manually. 
• Heartbeat.ai : if you know that people are talking about your market, it’s also important to know what they’re saying. Just because a lot of people are talking about Blockbuster, doesn’t mean DVD rental is back in the picture. They’re just making memes. With heartbeat.ai, you can automatically analyse the tone of online conversations. 

Love these tools? So do we. If you’re looking for more of this, check out our entire article with 6 of our favourite tools for market research . 

Track product demand with Attest

Learn how your customers think, act and buy to give them the products they want. With an audience of over 125 million people in 59 countries, our data gives you an accurate picture of your product demand.

VP Customer Success 

Sam joined Attest in 2019 and leads the Customer Research Team. Sam and her team support brands through their market research journey, helping them carry out effective research and uncover insights to unlock new areas for growth.

Related articles

11 customer profiling tools to understand your market, consumer profiling, how to conduct a brand tracking study, brand tracking, 7 brand intelligence software to collect insights, subscribe to our newsletter.

Fill in your email and we’ll drop fresh insights and events info into your inbox each week.

* I agree to receive communications from Attest. Privacy Policy .

You're now subscribed to our mailing list to receive exciting news, reports, and other updates!

Brought to you by:

Hind Oil Industries: Demand Analysis

By: Abhishek Rohit, Debdatta Pal, Pradyumna Dash

In September 2015, the manager of Hind Oil Industries (HOI), a small edible oil manufacturer in Asansol, West Bengal, was challenged by a dilemma in pricing strategy. HOI had seen the price of…

• Length: 10 page(s)
• Publication Date: Apr 20, 2017
• Discipline: General Management
• Product #: W17229-PDF-ENG

What's included:

• Teaching Note
• Educator Copy
• Supplements

$4.95 per student

degree granting course

$8.95 per student

non-degree granting course

Get access to this material, plus much more with a free Educator Account:

• Access to world-famous HBS cases
• Up to 60% off materials for your students
• Resources for teaching online
• Tips and reviews from other Educators

Already registered? Sign in

• Student Registration
• Non-Academic Registration
• Included Materials

In September 2015, the manager of Hind Oil Industries (HOI), a small edible oil manufacturer in Asansol, West Bengal, was challenged by a dilemma in pricing strategy. HOI had seen the price of mustard seeds, its primary raw ingredient, rise steeply because of harsh weather conditions in the previous harvesting season. HOI would need to raise the price of its product significantly in a price-competitive market dominated by larger companies. The manager wondered if he could increase the price of his only product, Maa mustard oil, to cover the substantial increase in production costs without suffering a loss in total revenue earned. If so, what would be the optimum price under various scenarios related to his competitors' expected price hikes? Could HOI's price be raised even if the competitors decided against raising their prices?

The authors are affiliated with Indian Institute of Management Raipur.

Learning Objectives

This case has been designed for senior undergraduate and graduate students in managerial economics and microeconomics courses dealing with quantitative demand analysis, demand estimation and forecasting, demand modelling, pricing decisions, and entrepreneurial decision-making for small organizations. After completion of this case, students will be able to apply quantitative demand analysis to a real business problem using a multiple regression technique; understand the various elasticities of demand and their implications; describe the concept of the total revenue test; apply the concepts of elasticities of demand and the total revenue test to a real business situation; calculate the optimum price for maximizing total revenue using the demand function; use a regression technique for forecasting demand; and analyze a scenario in order to make business decisions.

Apr 20, 2017

Discipline:

General Management

Geographies:

Ivey Publishing

W17229-PDF-ENG

We use cookies to understand how you use our site and to improve your experience, including personalizing content. Learn More . By continuing to use our site, you accept our use of cookies and revised Privacy Policy .

International Conference on Computational Science and Its Applications

ICCSA 2019: Computational Science and Its Applications – ICCSA 2019 pp 50–63 Cite as

Demand Forecasting: A Case Study in the Food Industry

• Juliana C. Silva   ORCID: orcid.org/0000-0003-2835-4196 18 ,
• Manuel C. Figueiredo   ORCID: orcid.org/0000-0002-0483-1341 19 &
• Ana C. Braga   ORCID: orcid.org/0000-0002-1991-9418 19  
• Conference paper
• First Online: 29 June 2019

2668 Accesses

3 Citations

Part of the book series: Lecture Notes in Computer Science ((LNTCS,volume 11621))

The use of forecasting methods is nowadays regarded as a business ally since it supports both the operational and the strategic decision-making processes.

This paper is based on a research project aiming the development of demand forecasting models for a company (designated here by PR) that operates in the food business, more specifically in the delicatessen segment.

In particular, we focused on demand forecasting models that can serve as a tool to support production planning and inventory management at the company.

The analysis of the company’s operations led to the development of a new demand forecasting tool based on a combination of forecasts, which is now being used and tested by the company.

This is a preview of subscription content, log in via an institution .

Buying options

• Available as PDF
• Read on any device
• Instant download
• Own it forever
• Available as EPUB and PDF
• Compact, lightweight edition
• Dispatched in 3 to 5 business days
• Free shipping worldwide - see info

Tax calculation will be finalised at checkout

Purchases are for personal use only

Acar, Y., Gardner, E.S.: Forecasting method selection in a global supply chain. Int. J. Forecast. 28 (4), 842–848 (2012). https://doi.org/10.1016/j.ijforecast.2011

Article   Google Scholar  

Bates, J.M., Granger, C.W.J.: The combination of forecasts. OR 20 (4), 451–468 (1969)

Bunn, D.W.: A Bayesian approach to the linear combination of forecasts. Oper. Res. Q. 26 (2), 325–329 (1975). (1970–1977)

Chan, C.K., Kingsman, B.G., Wong, H.: The value of combining forecasts in inventory management - a case study in banking. Eur. J. Oper. Res. 117 (2), 199–210 (1999). https://doi.org/10.1016/S0377-2217(98)00277-X

Article   MATH   Google Scholar  

Clemen, R.T.: Combining forecasts: a review and annotated bibliography. Int. J. Forecast. 5 (4), 559–583 (1989). https://doi.org/10.1016/0169-2070(89)90012-5

Diebold, F.X.: Serial correlation and the combination of forecasts. J. Bus. Econ. Stat. 6 (1), 105–111 (1988). https://doi.org/10.1080/07350015.1988.10509642

Article   MathSciNet   Google Scholar  

Gaither, N., Frazier, G.: Administração da Produção e Operações. Thomson Learning (2007)

Google Scholar  

Genre, V., Kenny, G., Meyler, A., Timmermann, A.: Combining expert forecasts: can anything beat the simple average? Int. J. Forecast. 29 (1), 108–121 (2013). https://doi.org/10.1016/j.ijforecast.2012

Granger, C.W.J., Ramanathan, R.: Improved methods of combining forecasts. J. Forecast. 3 (2), 197–204 (1984). https://doi.org/10.1002/for.3980030207

Holt, C.C., Modigliani, F., Muth, J.F., Simon, H.A., Bonini, C.P., Winters, P.R.: Planning Production, Inventories, and Work Force. Prentice-Hall, Upper Saddle River (1960)

Makridakis, S., et al.: The accuracy of extrapolation (time series) methods: results of a forecasting competition. J. Forecast. 1 (2), 111–153 (1982). https://doi.org/10.1002/for.3980010202

de Menezes, L.M., Bunn, D.W.: The persistence of specification problems in the distribution of combined forecast errors. Int. J. Forecast. 14 (3), 415–426 (1998)

Winters, P.R.: Forecasting sales by exponentially weighted moving averages. Manag. Sci. 6 (3), 324–342 (1960). https://doi.org/10.1287/mnsc.6.3.324

Article   MathSciNet   MATH   Google Scholar  

Wong, W., Guo, Z.: A hybrid intelligent model for medium-term sales forecasting in fashion retail supply chains using extreme learning machine and harmony search algorithm. Int. J. Prod. Econ. 128 (2), 614–624 (2010)

Download references

Acknowledgments

This work has been supported by FCT – Fundação para a Ciência e Tecnologia within the Project Scope: UID/CEC/00319/2019.

Author information

Authors and affiliations.

DPS, School of Engineering, University of Minho, Braga, Portugal

Juliana C. Silva

ALGORITMI Centre, University of Minho, 4710-057, Braga, Portugal

Manuel C. Figueiredo & Ana C. Braga

You can also search for this author in PubMed   Google Scholar

Corresponding author

Correspondence to Manuel C. Figueiredo .

Editor information

Editors and affiliations.

Covenant University, Ota, Nigeria

Sanjay Misra

University of Perugia, Perugia, Italy

Osvaldo Gervasi

University of Basilicata, Potenza, Italy

Beniamino Murgante

Saint Petersburg State University, Saint Petersburg, Russia

Elena Stankova

Vladimir Korkhov

Polytechnic University of Bari, Bari, Italy

Carmelo Torre

University of Minho, Braga, Portugal

Ana Maria A.C. Rocha

Monash University, Clayton, VIC, Australia

David Taniar

Kyushu Sangyo University, Fukuoka, Japan

Bernady O. Apduhan

Eufemia Tarantino

Rights and permissions

Reprints and permissions

Copyright information

© 2019 Springer Nature Switzerland AG

About this paper

Cite this paper.

Silva, J.C., Figueiredo, M.C., Braga, A.C. (2019). Demand Forecasting: A Case Study in the Food Industry. In: Misra, S., et al. Computational Science and Its Applications – ICCSA 2019. ICCSA 2019. Lecture Notes in Computer Science(), vol 11621. Springer, Cham. https://doi.org/10.1007/978-3-030-24302-9_5

Download citation

DOI : https://doi.org/10.1007/978-3-030-24302-9_5

Published : 29 June 2019

Publisher Name : Springer, Cham

Print ISBN : 978-3-030-24301-2

Online ISBN : 978-3-030-24302-9

eBook Packages : Computer Science Computer Science (R0)

Share this paper

Anyone you share the following link with will be able to read this content:

Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative

• Publish with us

Policies and ethics

• Find a journal
• Track your research

A forward-looking supply chain using demand forecasting

Tomorrow’s demands, forecasted today.

5-MINUTE READ

Call for change

Looking out to see within

A leader in food marketing and distribution saw an opportunity to reimagine its supply chain management.

Accenture introduced the idea that by using unified view of demand, the company could develop a supply chain that anticipates and pivots around obstacles.

The goal was to combine internal data with new external data that had emerged during the pandemic to gain  greater visibility and flexibility .

From there, the company could scale the new solution across all of its operational sites to better inform inventory position and supply-side dynamics, future-proofing the company’s operations and giving it an edge over competitors.

When tech meets human ingenuity

Stepping towards the future

Accenture proposed a five-week engagement to prove the value of unified view of demand by focusing on analytics and insights.

Accenture combined internal data (like sales and inventory) and external data (like weather and restaurant reservations) into  an AI-driven solution  that could easily forecast and improve demand sensing.

From there, Accenture replicated and expanded unified view of demand to additional sites while live piloting the solution against the company’s existing supply chain process and system. Following the new solution’s success, Accenture outlined the specific steps needed to implement unified view of demand, from design flow to staffing to project timeline and beyond.

A valuable difference

Test. Analyze. Refine. Repeat.

After piloting  unified view of demand  across several sites, the company discovered that it could  improve forecast errors by roughly 6-8 points, which could lead to $100-$130M in potential benefits .

Accenture also brought operational excellence to the table by introducing AI-enabled exception-based management.

Given the tens of thousands of stock-keeping units (SKUs) the company faces in each iteration, the proposed solution enables planners to focus solely on the SKUs with critical need, saving time and making the entire process more robust.

Now, the company’s leadership is looking for other areas to innovate which has led to a culture of continuous learning.

Today, the unified view of demand forecasting model is an AI-powered solution that can inform demand forecasting and better prepare the company for the future.

Demand and Supply: Case Study

• February 26, 2023

Download the dataset below to solve this Data Science case study on Analyzing Demand and Supply.

Analyzing Demand and Supply: Case Study

Understanding demand and supply is critical for businesses to optimize operations, maximize profits, and make informed decisions. Analyzing the demand and supply for cab ride businesses like Ola and Uber is one of the challenging use cases of demand and supply analysis. 

Cab services have become an essential part of urban transportation, with people relying heavily on these services for their daily commutes. Understanding the demand and supply patterns of cab services can help optimize their operations and provide a better user experience to customers.

Here is a dataset of the demand for rides and the supply of drivers in a particular city. Below are the features in the dataset:

• Drivers Active Per Hour: Number of drivers active per hour.
• Riders Active Per Hour: Number of Riders looking for rides.
• Rides Completed: Number of rides completed.

Your task is the analyze the demand for rides and the supply of cabs to understand demand and supply patterns.

References to Solve this Data Science Case Study

• Demand and Supply Analysis by Aman Kharwal

Recommended Case Studies & Datasets

Training Multiple Models: Case Study

• February 19, 2024

Dynamic Pricing: Case Study

• June 25, 2023

Retail Price Optimization: Case Study

• April 15, 2023

Discussion Cancel reply

Discover more from statso.

Subscribe now to keep reading and get access to the full archive.

Type your email…

Continue reading

Demand Forecasting

To increase the accuracy of our client's demand planning, we developed a demand forecasting engine that combines a wide variety of machine learning and deep learning algorithms to predict the demand for over 20,000 products within the next 24 months.

• Industry Automotive
• Topic Forecasting
• Tools Python, Tensorflow, Docker, Azure
• Duration 1.5 years

Demand planning is a central component of a smooth supply chain for the entire product range. In order to avoid inventory surpluses or supply bottlenecks, demand must ideally be precisely planned months or even years ahead of the delivery date. Our customer was faced with the challenge of manually planning several thousand products every month. This ties up a lot of resources, with the sheer volume making it difficult to take important external influencing factors into account. An automated forecasting engine was to remedy this situation, as historical patterns and external influencing factors can be used to accurately plan ahead demand on a monthly basis.

To support strategic demand planning, we have developed a forecasting engine that uses various machine learning and deep learning algorithms to project demand up to 24 months into the future. The engine automates data preparation, the selection of external influencing factors, model estimation, and model selection and combination. The solution combines different modeling approaches so that seasonality, trends and external influencing factors can be optimally estimated. In order to plan the complete product range, the engine was deployed in the cloud and forecasts are automatically transferred to target systems.

Based on the engine’s forecasts, the accuracy of demand predictions across the entire product range was increased by around 10 percent. In addition, the patterns of the forecasts are convincing, as they take seasonality, trends and external influencing factors into account and extrapolate into the future. The engine’s predictions were tested in other markets during the course of the project, where they were able to achieve improvements in accuracy of around 10 percent, even without being adapted to the specific market. The predictions can now be used to validate and use forecasts in strategic planning to significantly reduce monthly planning time.

Marcel Plaschke

Learn more.

• AI Strategy
• AI Development
• AI Ventures

Vielen Dank für Ihre Nachricht. Wir werden uns schnellstmöglich mit Ihnen in Verbindung setzen. Freundliche Grüße Ihr statworx Team

McDonald’s Supply Chain Issues – a Case Study on Supply and Demand Analysis

Is there a more iconic symbol of American culture than McDonald’s?

The fast-food industry has shaped the modern lifestyle, not only in the USA. McDonald’s Corporation is present in over 100 countries and has more than 1.7 million employees. There are a lot of helpful business lessons a student can learn from doing the McDonald’s case study .

• 🍔 McDonald’s Supply Chain
• 🥬 Case Study Ideas
• 🚚 Supply and Demand Analysis
• 📈 Supply Analysis
• 💊 Supply Chain Issues
• 📚 Demand Analysis
• 👍 McDonald’s Case Study – 30 Best Examples

🔗 References

🍔5 facts about mcdonald’s supply chain.

• A new McDonald’s opens nearly every 14.5 hours . McDonald’s supply chain management made the company the number one fast-food chain globally.
• McDonald’s has one of the best supply chains in the world . In 2016, it was ranked the second-best in the Top Supply Chains by Gartner.
• McDonald’s uses The Three-Legged Stool approach for franchising . The company’s founder Ray Kroc designed this method: the first leg is McDonald’s employees, the second is the franchise owners, and the third is the suppliers.
• Products are transported to McDonald’s restaurants by special cars . They are equipped with innovative temperature-control mechanisms. Check out our free essay samples to learn more about McDonald’s supply chain.

🥬 McDonald’s Case Study – Fresh Ideas

Various factors contribute to McDonald’s successful financial performance. Here’re some ideas to consider when doing the McDonald’s case study.

• Corporate culture.
• Vertical integration of the supply chain.
• International menu variations.
• McDonald’s inventory management system.
• Storage and distribution systems.
• Marketing principles.
• Recruitment practices at McDonald’s.
• Suppliers’ code.
• McDonald’s largest suppliers.
• The SWOT analysis .

🚚 McDonald’s Supply and Demand Analysis

• McDonald’s supply chain overview McDonald’s outsources 100% of its supply needs. The company grows its products through contracted producers and transports its own goods. McDonald’s supply chain promotes growth not only for the restaurants but also for the suppliers.
• McDonald’s consumers’ demands Most customers expect quick and simple ordering from McDonald’s restaurants. Having failed the experiment with pizza in the 1990s , McDonald’s has stuck to hamburgers, cheeseburgers, and French fries on the menu.
• McDonald’s model is a win-win relationship between the company, the suppliers, and the customers.
• Instead of looking for suppliers that offer the best price, McDonald’s maintains long-term relationships with existing ones.
• McDonald’s does not micromanage business partners in its business model.

📈 McDonald’s Supply Chain Case Study

• McDonald’s supply chain: sustainability McDonald’s aims to use natural resources to contribute to protecting the environment. One of the significant steps in this process is utilizing nature-positive supply chains. While trying to reduce their impact on nature, the company also develops a more sustainable global economy.
• McDonald’s supply chain: economics McDonald’s works with its suppliers to research and share the best practices to increase productivity and financial performance. The company aims to reduce hunger and provide decent jobs to its suppliers worldwide.
• McDonald’s supply chain: food safety Food safety is a serious responsibility of McDonald’s. To help with this, suppliers have special programs allowing them to track food in the supply chain. Moreover, McDonald’s approaches a Food Safety Advisory Council to educate its employees about food safety.

💊 McDonald’s Supply Chain Issues

• McDonald’s switching from frozen to fresh burgers Since 2012, McDonald’s has lost 500 million restaurant visits as customers have chosen more fast-casual alternatives that provide fresh burgers. For that reason, in 2018, the company announced that it was switching to fresh patties instead of frozen in its burgers at 300 U.S. restaurants.
• McDonald’s supply chain issues due to Covid-19 In 2021, McDonald’s lacked some bottled drinks across its 1,250 restaurants in Great Britain. This happened because of a shortage of lorry drivers due to Covid-19 restrictions and self-isolation rules.
• McDonald’s supply chain disruptions Supply chain disruptions can significantly impact McDonald’s financial performance and reputation. To reduce the negative impact, McDonald’s has multiple supply backup sources. McDonald’s has tested and trustworthy suppliers ready to help if a disruption occurs.

📚 McDonald’s Demand Analysis

• Menu and burgers at McDonald’s McDonald’s used to make the tastiest hamburgers in America. However, this award today goes to other fast-food restaurants, such as Shake Shack . To win back customers, McDonald’s needs to focus on improving the quality of its core products.
• Lower prices at McDonald’s McDonald’s customers are unwilling to spend more than $5 on a hamburger. McDonald’s has a simplified menu and self-serve ordering system to lower its prices. It is important to notice that organic food and high quality are not the first things consumers want from McDonald’s.
• Customer service at McDonald’s McDonald’s employees’ essential quality is having a passion for people. McDonald’s adopts a multi-channel communication system to ensure that all its messages from the customers are delivered to staff. McDonald’s collects feedback and complaints on its website to identify factors affecting the demand .

👍 McDonald’s Case Study – 27 Best Examples

Mcdonald’s supply chain issues.

• Food industry: McDonald’s company . This essay investigates McDonald’s price elasticity of demand for its goods and services and its supply chain.
• Mcdonald’s vs. Wal-Mart’s Strategic Choices . This paper compares the two successful American companies and their supply chains.
• McDonald’s organization: operation management . This report focuses on the key concepts of McDonald’s operations management, including supply chain management.
• Mcdonald’s entering Estonia case analysis . This study investigates how McDonald’s can enter the Estonian market successfully and build a trustworthy relationship with local suppliers.
• An external and internal analysis of McDonald’s Corporation . This paper conducts an external and internal analysis of McDonald’s Corporation and evaluates its supply chain management.
• Quality management in McDonald’s restaurant . This essay analyses the measures McDonald’s uses to ensure that all supply chain operations occur safely.
• Business strategy analysis of Mcdonald’s . This paper explores how McDonald’s supply chain supports its operations worldwide.
• McDonald’s and Wendy’s International Inc . This report focuses on the management, suppliers, and customers of the two competing companies— McDonald’s and Wendy’s.
• McDonald’s, Starbucks, and American International Group . This paper compares three selected public companies, focusing on their financial performances and supply chains.

McDonald’s Marketing Strategies & Advertisement Campaign

• McDonald’s and Coca-Cola ads in the Russian market . This essay explores McDonald’s and Coca-Cola’s ads’ impact on the Russian audience.
• Marketing analysis: McDonald’s company. This study explores McDonald’s marketing approach to maintaining the balance between price and quality.
• McDonald’s marketing strategies in the UAE. The report analyzes the marketing strategies of McDonald’s, as well as market descriptions and segments of the company in the UAE.
• McDonald’s company’s marketing strategies . This report presents an analysis of the internal and external environments of McDonald’s and reviews its marketing mix.
• McDonald’s company marketing practices and strategies . This paper looks at the most suitable market entry strategies McDonald’s can use to continue expanding to the international market.
• Business ethics and Social Corporate Responsibility for McDonald’s . This essay investigates how McDonald’s CRS policy contributes to the company’s marketing success.

McDonald’s Human Resource Management & Employment

• McDonald’s business principles: employment violations . This essay concerns the problem of the company’s ethics strategy and how McDonald’s tries to fix it.
• McDonald’s strategic management of human resources & innovation . This research contains an overview of McDonald’s human resource management and suggests how it can be improved.
• McDonald’s human resources management practice . This paper explores McDonald’s human resource management and its strengths and weaknesses.
• McDonald’s company: human resource functions. This research focuses on McDonald’s human resource issues.
• McDonald’s company H.R. management practices . This essay aims to demonstrate how McDonald’s efficient H.R. management practices benefit the company’s financial performance.
• McDonald’s Corporation’s talent management program . This paper provides details on McDonald’s successful talent management program.

Mcdonald’s Financial Performance & Market Influence

• McDonald’s company case analysis . This paper aims to assess the effects of McDonald’s operations strategy and its different perspectives on the company’s financial success.
• McDonald’s Corporation’s five forces analysis. This essay focuses on the five internal and external factors influencing McDonald’s performance.
• Microsoft Corporation and McDonald’s corporation: financial performance . This paper compares Microsoft Corporation and McDonald’s Corporation’s financial performance.
• McDonald’s accounting information system . This paper examines the accounting information systems at McDonald’s.
• McDonald’s company’s strategic management. This essay investigates how McDonald’s strategic decisions influence the company’s financial performance.
• Comparative financial statements of McDonald’s. This writing analyzes the financial aspects of McDonald’s over the recent years.
• Best Items on McDonald’s All Day Breakfast Menu – Business Insider
• The Supply Chain: From Raw Materials to Order Fulfillment – Investopedia
• 2022: Supply chains will face many challenges this year | World Economic Forum
• McDonald’s Is Using AI and Data to Optimize Its Supply Chain
• Demand and Supply Planning for a Large Fast Food Chain – AnyLogic Simulation Software

Descriptive Essay Topics: Examples, Outline, & More

371 fun argumentative essay topics for 2024.

The Economics at Play

Exploring the relevance of economic theory in today's world

Supply and Demand – A Case Study on LNG

Why you should read this article:

• Energy prices in Europe have reached record-breaking levels in the past months
• This is specifically true when looking at one of the bloc’s most important sources of energy: natural gas
• What has caused Europe’s natural gas prices to skyrocket so drastically? In short, supply and demand
• In this article, factors affecting both sides are be explored to better understand how the market forces at play affect the pockets of everyday customers

Throughout Europe, markets have seen a notable surge in the price of energy. This is mostly attributed to the global situation surrounding one of the staple energy sources in the EU bloc: natural gas. For instance, liquified natural gas (LNG) , which arrives at European ports through shipping vessels to then be turned back into a gas at regasification plants, reached a price of about EUR 93 per megawatt- hour on September 26th, 2021  (compared to only EUR 13 one year prior). With winter looming around the corner, many policymakers are becoming increasingly worried as to how these high energy prices will impact both households and businesses in the form of increased electricity bills.  

Tight supply, coupled with increased demand, have created a unique situation in which energy prices, specifically LNG prices, have skyrocketed to levels unseen before in the area. The price of LNG , on September 26 th , reached about EUR 93 per megawatt-hour (compared to EUR 13 one year prior). What has caused Europe’s natural gas prices to skyrocket so drastically? In short, supply and demand.

Figure 1: Natural Gas Prices in Europe

While the situation is dynamic and complicated, it can be boiled down and explained, in simple terms, as a problem of basic economics. In this article, factors affecting both sides will be explored, within the context of Europe, to better understand how the market forces at play affect the pockets of everyday customers. 

Demand –  The consumer’s side of the coin 

Colder winter, warm summer

The last months of 2020 experienced a colder-than-expected winter in Europe, which pushed demand past previously forecasted levels. In conjunction with the Covid-19 pandemic-related measures which forced many workers to work from home, households found themselves spending even more energy to keep homes comfortable. Already in January, LNG was selling at levels higher than in January 2019 . This increase in demand forced producers to take out gas from their storage reserves because the production supply was not sufficient to meet demand.

After winter ended and the second quarter of the year began, consumers’ needs shifted from heating to cooling and offered no respite to natural gas demand levels. Heatwaves across Europe saw demand for energy increase as people sought to cool off indoors. During this quarter, European gas consumption increased by 25%, the largest year-on-year quarterly increase since 1985 . 

Post-lockdown spending

During the coronavirus pandemic, households in the eurozone saw their savings increase (on aggregate) to 20% of the nominal disposable income in 2020 (up from 13.1% in 2019). With the easement of lockdowns across the bloc, consumers were ready to spend. In May 2021, the EU saw an increase in consumer spending of 17.8%. The increase in consumption increased the price of energy commodities, particularly natural gas (and LNG), as they signaled a recovery of economic activity in the area, which is supported by energy consumption. 

The ETS effect

On the industrial side of demand, producers saw the price of an ETS allowance increase past expectations. Due to this increase in the price of ETS credits, industrial players were incentivized to switch from carbon-intensive fuels (such as coal) to natural gas. This incentive was primarily driven by a desire to minimize costs that are associated with the purchase of emission allowances. In turn, as industrial players aimed to minimize costs, their demand for natural gas (and LNG) also increased.

Supply – How much pie is there to go around? 

Low storage levels

Normally, natural gas producers take advantage of low-demand summer months to ramp up their underground storage supplies in order to prepare for the (usually) more demand-intensive winter months. Having these reserves allows producers to have a backup with which to operate in the face of demand fluctuations. Given the demand factors studied above, producers in the area had no such opportunities this past summer in 2021. Low storage levels ensued. Already in March 2021, low storage levels were creating worries. At 30% , levels were historically low for the time of year. By September 2021, EU-wide levels were reported at 25% less than their long term average as they head into high-demand winter months. 

The result: Lower-than-ideal storage levels to deal with high demand in a potentially cold winter, creating upwards pressure on prices. 

Figure 2: Natural gas inventory levels in the EU

The EU depends heavily on natural gas imports. In 2020, natural gas represented approx. 25% of total primary energy consumption in the EU, with the bloc importing most of it. The group has two major importers, Russia and Norway, both of which have witnessed supply interruptions in the past year. Amongst other factors at play, this led to lower supply levels than expected. 

Adding more fuel to the supply “fire”, the EU also competes with other regions for LNG imports, the most notable of which is Asia. As a region, Asia also experienced several factors (incl. cold winters of its own), which increased demand in the region. With countries such as Japan phasing out nuclear power, the role LNG imports played in these economies became increasingly important. The result was a trend in which suppliers sent LNG exports to Asia, instead of to the EU, as the buyers in the Asia region offered to pay higher prices . This persisted even as prices increased inside of the eurozone. 

The result: Decreased supply of LNG in the form of imports, with demand levels that do not decrease, resulting in an increase in the price of LNG traded in the EU bloc.

Insufficient availability of other sources in the fuel mix 

As LNG supplies grew more limited, typical fuel switching (substituting power generation from one energy source to another) in Europe was not straightforward to achieve. Governments, seeking to decarbonize their economies, had reduced coal and oil power production and had switched to natural gas to minimize ETS related costs. On the other hand, renewable generation sources, particularly wind and solar, did not contribute enough power generation to alleviate the supply crunches observed with LNG. Nuclear power is not a particularly popular opinion in the bloc, and hence not a feasible solution to the supply shock.

The result: Lack of LNG supplies not relieved by fuel switching, offering no relief to price levels. 

Keeping the lights on

With the market forces described above, where does that leave the EU? Already, worries regarding inflation have come afloat. The Euro reported its inflation to have increased to 3.4% in September 2021 (compared to August 2021). According to the EU’s Statistical Office, the main component causing the rise was energy, which experienced an increase of 17.4% in September. 

Furthermore, fears of surging consumer electricity bills and factory shut downs have fueled calls for government action. Some governments have already started to do so. Spain has introduced a measure to cap gas prices for consumers. Italy announced multi-billion euro measures to ease energy bills of households in the last three months of 2021. But for now, these actions are temporary solutions to a much more complicated and fragile situation, whose outcome will remain to be seen. Undoubtedly, policymakers will be regularly checking weather forecasts. If the upcoming winter is long and cold, the demand and supply factors at play will get stronger, leaving ordinary consumers to pay for the consequences with their pockets. 

You might also like

What Drives Wallstreet? A Quantitative Analysis of Institutional Trading

European EUAs and the COVID Shock

Voting: How the Result is Predetermined by the Election System

Terms and Conditions - Privacy Policy

Want to create or adapt books like this? Learn more about how Pressbooks supports open publishing practices.

Note that the Economics 103 Case Studies are meant to supplement the course material by giving you experience applying Economic concepts to real world examples. While they are beyond the level you will be tested on, they are useful for students who want a stronger grasp of the concepts and their applications.

Note that this case study is difficult if you do not print the diagrams, or reproduce them on graph paper. If you are unable to print, we recommend reviewing the solutions to ensure you understand the general lessons presented.

In 2016 rental vacancy rates dropped to as low as 0.6% in Greater Victoria. When compared to the national average of 3.3% it is clear why many media channels and individuals were calling it a ‘housing crisis’. Students were especially hit hard by these low vacancies, with some international students at Camoson college having to return home when they couldn’t find a place to stay. Using our competitive market model, let’s examine some of the factors that played into this crisis and policies that could be used to fix it.

Read more about the Victoria Housing crisis.

Below is a representation of the Victoria Housing Market.

1. Label Figure CS3 a. with the Equilibrium price and quantity, and label supply and demand curves as either renters or landlords.

If supply is equal to demand there should be no vacancy, but we know that a 0% vacancy rate would be an extremely difficult market for renters.

2. Explain why a housing market at equilibrium could still have a vacancy rate of 4%.

One factor that has been blamed for the housing crisis is Airbnb. Airbnb describes itself as  an online marketplace and hospitality service, enabling people to list or rent short-term lodging including vacation rentals, apartment rentals, homestays, hostel beds, or hotel rooms, with the cost of lodging set by the property owner. City councillors have targeted these short term rentals, saying that many landlords have opted to Airbnb their home, rather than rent out longer term. The growth of Airbnb in Vancouver has been shown below.

3. Assume 3000 landlords decide to switch from renting to Airbnb, show the impact of the changes on Figure CS3 b. Label the new equilibrium price and quantity.

Note that Airbnb has been adamant that short-term rentals have had a neglible impact on the housing market, citing that in Vancouver only 320 hosts rent out thier properties often enough to make more money that long term rentals. That represents only 0.11% of the total housing units. In Victoria, that would mean only 25 units are affected by short-term rentals.

Tom Davidoff , a  University of B.C. business professor, said t he general public frequently looks at the fact that Airbnb is popular in expensive neighbourhoods and concludes that it is Airbnb that drives up rents there. But, he said, those neighbourhoods were expensive anyway and the impact of Airbnb taking a certain slice of the available stock is minimal.

Read more about Airbnb’s supposed impact on the market.

Another factor that has had an impact on the rental market is the inability of many young Canadians to buy homes. Not only have house prices skyrocketed, but more are burdened by student loans out of university. It is estimated that it takes 3 times longer (15 years) to save enough to have a 20% downpayment on a house than it did in 1976. This reduction in Canadians mortgaging a home has caused an increasing amount to enter the market.

Read more about the increasing squeeze on millennials.

4. Assume 9000 new renters enter the market instead of mortgaging homes, show the impact of the changes on Figure CS3 b.  Explain the impact of both the shock from Airbnb and the shock from less housing buyers on equilibrium price and quantity. Do the shocks work together or oppose one another?

In the housing market, prices are slow to adjust, landlords cannot simply raise prices immediately under the Residential Tenancy Act. Landlords can only raise prices when negotiating a new contract. This causes many unjustified evictions from landlords as they want to charge the new equilibrium price. In the short-term, prices stay relatively the same causing shortages or surplus.

5. Assume price remains at the original equilibrium , calculate the magnitude of the shortage or surplus of housing that results. Explain the impact this shortage will have the behaviour of landlords.

The British Columbia government unveiled a $500-million affordable-housing plan targeted at communities that have struggled with a shortage of low-cost housing.  Premier Christy Clark announced her government’s commitment to fund 68 new projects to help address the crisis

Read more about the governments response to the housing crisis.

6. Assume the government wants to bring price back to it’s original level, if it costs $50,000 to increase the number of rental units by one, how much will this cost the government?

Groups have criticized the government response, saying that they have ignored many other avenues that could more easily increase the supply of affordable housing.

7. Read the Executive Summary of the Alliance of BC Students White Paper on Student Housing.

What is the ABCS proposing that could help decrease price in the market? How would this affect supply and/or demand?

In this case study we have shown how microeconomic concepts of supply and demand can be used to understand current events in the news. Do you have a story you think would make a good case study? Contact [email protected] to propose your story.

Principles of Microeconomics Copyright © 2017 by University of Victoria is licensed under a Creative Commons Attribution 4.0 International License , except where otherwise noted.

Share This Book

We use cookies to enhance our website for you. Proceed if you agree to this policy or learn more about it.

• Essay Database >
• Essays Samples >
• Essay Types >
• Case Study Example

Demand Case Studies Samples For Students

496 samples of this type

Do you feel the need to check out some previously written Case Studies on Demand before you begin writing an own piece? In this open-access collection of Demand Case Study examples, you are provided with a thrilling opportunity to examine meaningful topics, content structuring techniques, text flow, formatting styles, and other academically acclaimed writing practices. Exploiting them while crafting your own Demand Case Study will surely allow you to finish the piece faster.

Presenting high-quality samples isn't the only way our free essays service can aid students in their writing efforts – our experts can also create from point zero a fully customized Case Study on Demand that would make a genuine foundation for your own academic work.

Sample Case Study On Referring To The Article:

What is a cartel (5 marks), free price and quantity of milk case study sample, case study on economic, introduction.

Don't waste your time searching for a sample.

Get your case study done by professional writers!

Just from $10/page

Sample Case Study On Government Expenditures And Revenues

This paper is on the topic of Government Expenditures and Revenues, Budget, National Debt. The first part of this paper will discuss the effect of government purchases on aggregate demand. The second section of this paper will discuss the shift in aggregate demand curve and change in GDP in comparison to aggregate demand curve, on the bases of factors like government purchases, GDP and the multiplier. The Final section of this paper will discuss the issue of crowding out and its implications on fiscal policy changes.

Government Expenditures and Revenues, Budget and National Debt

Free case study on the variable costs of producing one unit = labor cost per hour + additional material cost, sunset grill at blue case studies example, executive summary.

Although Sunset Grill at Blue is experiencing a remarkable growth rate, managing demand during the peak periods such as weekends is still a major challenge. Customers have to wait for long to the extent that some resort to alternative restaurants. The management, therefore, is considering various ways of increasing service efficiency to reduce the queues and maintain its large customer base. After analyzing various options available, this study recommends that Sunset Grill at Blue should adopt a policy of hiring part-time workers and alternative services to help in enhancing service delivery and reducing the queuing time on weekends.

Free Case Study On Analyzing News Article In Microeconomics Terms

Case study on impact of ban on advertising ban on public smoking and display of risk warning on, free stage 1 case study: top-quality sample to follow, business cycle case study, sample case study on economics, example of premium pricing method case study, free the global and american biofuel situation2 case study sample, introduction2.

The opportunities and challenges of the biofuel industry..3

The demand of farm and food products by biofuel4

Inspiring case study about strategic analysis, strategic analysis alacer gold corp and gold and mining industry, free case study on the impact of product life cycle on supply chain strategy.

1. Integration of supply chain activities provides the competitive advantage to the company and facilitate company to fulfill changing demands of the customers. Opportunity of the growth and profitability resides within the supply chain of the company. This statement has proven not only once but in several cases. Efficient supply chain is directly linked with the quality of products and services that companies are offering. In competitive business environment, customers want companies to be responsive towards their needs, which can only be achieved with effective supply chain management.

Quotas: The Restrictions Imposed On The Amount Of A Production, Imports And Exports Case Study Examples

Chapter 8: tariffs versus quotas.

INTRODUCTION The main concern of this study is to determine the better option between quotas and tariff for US for importing goods from Japan. The impacts of both quota and tariff on the US economy along with the impact of all foreign imports on the domestic production in terms of industry competition is central idea of the study. Tariff: The amount of tax or duty that is to be paid to the government on the particular class of imports and exports.

Free Case Study On Review Of GIS In Housing And The Effect Of Educational Level

Case study on metabical case study, metabical: pricing, packaging, and demand forecasting for a new weight-loss drug, free analyzing article: the pot business suffer growing pains case study example, free case study on what is market failure.

Market failure occurs when the market does not clear due to lack of an equilibrium point. In most cases, it is when the supply does not meet the demand. It also occurs when there is an alternative situation for the same market to be in and more people to gain from it rather than that situation. The resources can be utilised for greater consumer utility.

What are the major reasons that a free unregulated market in medical care might not be optimal?

Example of case study on the operations of 7 eleven, free case study about relative price comparison, the impact of it on logistics integration case study examples, a case study of sunsweet growers.

The impact of IT on logistics integration – a case study of Sunsweet Growers

Case Study On Industrial Organization

Landmark antitrust case study assignment.

Dr. Mozayeni

Draw Topic & Writing Ideas From This Case Study On Project Progress

Ieg 330: supply chain & logistics engineering.

Section: F2 Summer 2016

Free Managing The Supply Chain Of Kuwait Indo Trading Company (Kitco) Case Study Example

Inspiring case study about tesla motors case study.

Enhancing performance requires effective identification of key issues that a business faces. That is because it helps in selection of suitable strategies as well as allocation of resources. With that, the following is a summary of the Tesla case study identifying the issues facing the company from the Market line case study and proposing suitable solutions as well as their implementation.

Comparing Mrp And Kanban Case Study Example

Material Requirements Planning (MRP) and Kanban control systems are one of the most prominent production planning and inventory control systems of today. These systems are the most popular implementations of push and pull strategies respectively. Speaking of MRP, it is a material control system that tries to maintain adequate inventory levels to ensure that the required materials are available, when needed. MRP is used where there are multiple items with complex bills of materials. It is not suited for continuous processes that are tightly linked.

The major objective of an MRP system is to simultaneously:

Case study on transportation economics case study of westminster company, price ceiling case studies example, example of doing business in africa case study, choosing a product, economics case study, relationship between consumption and gdp, example of case study on changing mindset case study analysis, good general practitioners case study example.

As Human Resources Manager, Care Concerns Pty Ltd (CC), explain the implications of the level of supply and demand in the labour market in the recruitment and retention of;

Good Example Of Logistics Planning Paper For Olive Oil Case Study

Logistics planning paper for olive oil 2, free exchange: macro control, micro problems case study sample, economic theory in the real world case study examples.

The media report presents the effects that the policy that the Australian government has taken. According to the media, the government has raised the minimum wage payment to the workers to $15 per hour.

Highlight the reason for the issue being raised

Example of case study on economics of japanese cars after the earthquake, equilibrium price and quantity, a-level case study on whirlpool supply chain for free use, background of the case, case study on operations management: just in time distribution, good example of case study on united states district for the district of columbia.

The United States district court for the District of Columbia is a case that was brought against Anheuser-Busch InBev SA/NV (ABI) to oppose the acquisition of Modelo group by ABI. It was brought up since the government felt that the proposed merger was violating the law through illegal ownership or acquisition of another firm. In addition, the government felt that the acquisition was threatening fair competition in a free market, and the consumers would suffer most as a result of the merger.

Facts to the Case

Competitive advantage case study, free case study on value model, de beers strategic case study case study examples.

1) As part of your environmental analysis, please identify the various components of the Global Diamond Industry (e.g., legal, regulatory, political, cultural, technological, and economic), highlighting the main trends affecting the future of this product market (up to 2 pages);

Exemplar Case Study On Whirlpool Supply Chain To Write After

Describe the challenges faced by the company. what were the drivers for change to the supply chain, case study on supply chain, strategic audit case study sample, written strategic audit.

(For The Recalcitrant Director at Byte Products, Inc. Case) I. Current Situation - Current Performance Byte has been a market leader for the past 6 years with steady annual income of 12% and a total turnover of about $265 million. Byte also has a market share of 32 percent. However, upsurge in demand has attracted entry of new firms placing its market leadership in jeopardy.

B. Strategic Posture

Good case study on space age furniture company, good case study about general graph for case 1, good case study about current markets conditions competitive analysis, instititution:, free case study about calculations.

The production director of Wilmslow Ltd wants to schedule production for quarter 3 (the twelve weeks ending 17 September) and asks you to use the revised information to prepare the following: - The revised production budget for Alphas and Betas;

Closing stock is 5 days and 10 days of sales for Q4, for alpha and beta respectively, which is 20% more than the new estimated demand, so the Q4 estimated demand is 120% of 2000 and 2400 respectively =120/100*2000=2400 2400/12=200 units (12 weeks because each week is 5 days)..For Alpha

Case Study On Tourism In Greece

Management case study sample, analyzing case study about adolph coors in the brewing industry, robert mondavi corporation case study, company overview, good sun edison company (sune) – strategic analysis case study example.

out from the fitted line. Now, tliers that spreads out from the fitted line. Now, in R&D with the symbol s

Consumption Expenditure And Gasoline Price Case Study Example

Consumption expenditure and gasoline price, new balance athletic shoe inc case study samples.

The case is about New Balance Athletic Shoe Inc. and how they are planning to increase their efficiency and quality in their operation processes through the implementation of lean manufacturing. At this same period, two of the leading shoe producers Reebok and Adidas have merged hence New Balance needs to focus on ways to counter this move in order to sell effectively.

Example Of Case Study On Studio Stessia

Anti-trust law case study examples, apple anti-trust case, example of customer-based production case study, implement lean elements on project management, netflix in 2012 (case in ebook) case study samples.

Password recovery email has been sent to [email protected]

Use your new password to log in

You are not register!

By clicking Register, you agree to our Terms of Service and that you have read our Privacy Policy .

Now you can download documents directly to your device!

Check your email! An email with your password has already been sent to you! Now you can download documents directly to your device.

or Use the QR code to Save this Paper to Your Phone

The sample is NOT original!

Short on a deadline?

Don't waste time. Get help with 11% off using code - GETWOWED

No, thanks! I'm fine with missing my deadline

• Open access
• Published: 17 April 2024

A data-driven combined prediction method for the demand for intensive care unit healthcare resources in public health emergencies

• Weiwei Zhang 1 &
• Xinchun Li 1  

BMC Health Services Research volume  24 , Article number:  477 ( 2024 ) Cite this article

Metrics details

Public health emergencies are characterized by uncertainty, rapid transmission, a large number of cases, a high rate of critical illness, and a high case fatality rate. The intensive care unit (ICU) is the “last line of defense” for saving lives. And ICU resources play a critical role in the treatment of critical illness and combating public health emergencies.

This study estimates the demand for ICU healthcare resources based on an accurate prediction of the surge in the number of critically ill patients in the short term. The aim is to provide hospitals with a basis for scientific decision-making, to improve rescue efficiency, and to avoid excessive costs due to overly large resource reserves.

A demand forecasting method for ICU healthcare resources is proposed based on the number of current confirmed cases. The number of current confirmed cases is estimated using a bilateral long-short-term memory and genetic algorithm support vector regression (BILSTM-GASVR) combined prediction model. Based on this, this paper constructs demand forecasting models for ICU healthcare workers and healthcare material resources to more accurately understand the patterns of changes in the demand for ICU healthcare resources and more precisely meet the treatment needs of critically ill patients.

Data on the number of COVID-19-infected cases in Shanghai between January 20, 2020, and September 24, 2022, is used to perform a numerical example analysis. Compared to individual prediction models (GASVR, LSTM, BILSTM and Informer), the combined prediction model BILSTM-GASVR produced results that are closer to the real values. The demand forecasting results for ICU healthcare resources showed that the first (ICU human resources) and third (medical equipment resources) categories did not require replenishment during the early stages but experienced a lag in replenishment when shortages occurred during the peak period. The second category (drug resources) is consumed rapidly in the early stages and required earlier replenishment, but replenishment is timelier compared to the first and third categories. However, replenishment is needed throughout the course of the epidemic.

The first category of resources (human resources) requires long-term planning and the deployment of emergency expansion measures. The second category of resources (drugs) is suitable for the combination of dynamic physical reserves in healthcare institutions with the production capacity reserves of corporations. The third category of resources (medical equipment) is more dependent on the physical reserves in healthcare institutions, but care must be taken to strike a balance between normalcy and emergencies.

Peer Review reports

Introduction

The outbreak of severe acute respiratory syndrome (SARS) in 2003 was the first global public health emergency of the 21st century. From SARS to the coronavirus disease (COVID-19) pandemic at the end of 2019, followed shortly by the monkeypox epidemic of 2022, the global community has witnessed eight major public health events within the span of only 20 years [ 1 ]. These events are all characterized by high infection and fatality rates. For example, the number of confirmed COVID-19 cases worldwide is over 700 million, and the number of deaths has exceeded 7 million [ 2 ]. Every major public health emergency typically consists of four stages: incubation, outbreak, peak, and decline. During the outbreak and transmission, surges in the number of infected individuals and the number of critically ill patients led to a corresponding increase in the urgent demand for intensive care unit (ICU) medical resources. ICU healthcare resources provide material security for rescue work during major public health events as they allow critically ill patients to be treated, which decreases the case fatality rate and facilitates the prevention and control of epidemics. Nevertheless, in actual cases of prevention and control, the surge in patients has often led to shortages of ICU healthcare resources and a short-term mismatch of supply and demand, which are problems that have occurred several times in different regions. These issues can drastically impact anti-epidemic frontline healthcare workers and the treatment outcomes of infected patients. According to COVID-19 data from recent years, many infected individuals take about two weeks to progress from mild to severe disease. As the peak of severe cases tends to lag behind that of infected cases, predicting the changes in the number of new infections can serve as a valuable reference for healthcare institutions in forecasting the demand for ICU healthcare resources. The accurate forecasting of the demand for ICU healthcare resources can facilitate the rational resource allocation of hospitals under changes in demand patterns, which is crucial for improving the provision of critical care and rescue efficiency. Therefore, in this study, we combined a support vector regression (SVR) prediction model optimized by a genetic algorithm (GA) with bidirectional long-short-term memory (BILSTM), with the aim of enhancing the dynamic and accurate prediction of the number of current confirmed cases. Based on this, we forecasted the demand for ICU healthcare resources, which in turn may enable more efficient resource deployment during severe epidemic outbreaks and improve the precise supply of ICU healthcare resources.

Research on the demand forecasting of emergency materials generally employs quantitative methods, and traditional approaches mainly include linear regression and GM (1,1). Linear regression involves the use of regression equations to make predictions based on data. Sui et al. proposed a method based on multiple regression that aimed to predict the demand for emergency supplies in the power grid system following natural disasters [ 3 ]. Historical data was used to obtain the impact coefficient of each factor on emergency resource forecasting, enabling the quick calculation of the demand for each emergency resource during a given type of disaster. However, to ensure prediction accuracy, regression analysis needs to be supported by data from a large sample size. Other researchers have carried out demand forecasting for emergency supplies from the perspective of grey prediction models. Li et al. calculated the development coefficient and grey action of the grey GM (1,1) model using the particle swarm optimization algorithm to minimize the relative errors between the real and predicted values [ 4 ]. Although these studies have improved the prediction accuracy of grey models, they mainly involve pre-processing the initial data series without considering the issue of the excessively fast increase in predicted values by traditional grey GM (1,1) models. In emergency situations, the excessively fast increase in predicted values compared to real values will result in the consumption of a large number of unnecessary resources, thereby decreasing efficiency and increasing costs. As traditional demand forecasting models for emergency supplies have relatively poor perfect order rates in demand analysis, which result in low prediction accuracy, they are not mainstream.

At present, dynamic models of infectious diseases and demand forecasting models based on machine learning are at the cutting edge of research. With regard to the dynamic models of infectious diseases, susceptible infected recovered model (SIR) is a classic mathematical model employed by researchers [ 5 , 6 , 7 ]. After many years of development, the SIR model has been expanded into various forms within the field of disease transmission, including susceptible exposed infected recovered model (SEIR) and susceptible exposed infected recovered dead model (SEIRD) [ 8 , 9 ]. Nevertheless, with the outbreak of COVID-19, dynamic models of infectious diseases have once again come under the spotlight, with researchers combining individual and group variables and accounting for different factors to improve the initial models and reflect the state of COVID-19 [ 10 , 11 , 12 , 13 ]. Based on the first round of epidemic data from Wuhan, Li et al. predicted the time-delay distributions, epidemic doubling time, and basic reproductive number [ 14 ]. Upon discovering the presence of asymptomatic COVID-19 infections, researchers began constructing different SEIR models that considered the infectivity of various viral incubation periods, yielding their respective predictions of the inflection point. Based on this, Anggriani et al. further considered the impact of the status of infected individuals and established a transmission model with seven compartments [ 15 ]. Efimov et al. set the model parameters for separating the recovered and the dead as uncertain and applied the improved SEIR model to analyze the transmission trend of the pandemic [ 16 ]. In addition to analyzing the transmission characteristics of normal COVID-19 infection to predict the status of the epidemic, many researchers have also used infectious disease models to evaluate the effects of various epidemic preventive measures. Lin et al. applied an SEIR model that considered individual behavioral responses, government restrictions on public gatherings, pet-related transmission, and short-term population movements [ 17 ]. Cao et al. considered the containment effect of isolation measures on the pandemic and solved the model using Euler’s numerical method [ 18 ]. Reiner et al. employed an improved SEIR model to study the impact of non-pharmaceutical interventions implemented by the government (e.g., restricting population movement, enhancing disease testing, and increasing mask use) on disease transmission and evaluated the effectiveness of social distancing and the closure of public spaces [ 19 ]. These studies have mainly focused on modeling the COVID-19 pandemic to perform dynamic forecasting and analyze the effectiveness of control measures during the epidemic. Infectious disease dynamics offer good predictions for the early transmission trends of epidemics. However, this approach is unable to accurately estimate the spread of the virus in open-flow environments. Furthermore, it is also impossible to set hypothetical parameters, such as disease transmissibility and the recovery probability constant, that are consistent with the conditions in reality. Hence, with the increase in COVID-19 data, this approach has become inadequate for the accurate long-term analysis of epidemic trends.

Machine learning has shown significant advantages in this regard [ 20 , 21 ]. Some researchers have adopted the classic case-based reasoning approach in machine learning to make predictions. However, it is not feasible to find historical cases that fully match the current emergency event, so this approach has limited operability. Other researchers have also employed neural network training in machine learning to make predictions. For example, Hamou et al. predicted the number of injuries and deaths, which in turn were used to forecast the demand for emergency supplies [ 22 ]. However, this approach requires a large initial dataset and a high number of training epochs, while uncertainty due to large changes in intelligence information can lead to significant errors in data prediction [ 23 , 24 , 25 ]. To address these problems, researchers have conducted investigations that account (to varying degrees) for data characterized by time-series and non-linearity and have employed time-series models with good non-linear fitting [ 26 , 27 , 28 ]. The use of LSTM to explore relationships within the data can improve the accuracy of predicting COVID-19 to some extent. However, there are two problems with this approach. First, LSTM neural networks require extremely large datasets, and each wave of the epidemic development cycle would be insufficient to support a dataset suitable for LSTM. Second, neural networks involve a large number of parameters and highly complex models and, hence, are susceptible to overfitting, which can prevent them from achieving their true and expected advantages in prediction.

Overall, Our study differs from other papers in the following three ways. First, the research object of this paper focuses on the specific point of ICU healthcare resource demand prediction, aiming to improve the rate of critical care patient treatment. However, past research on public health emergencies has focused more on resource prediction , such as N95 masks, vaccines, and generalized medical supplies during the epidemic , to mitigate the impact of rapid transmission and high morbidity rates. This has led to less attention being paid to the reality of the surge in critically ill patients due to their high rates of severe illness and mortality.

Second, the idea of this paper is to further forecast resource needs based on the projected number of people with confirmed diagnoses, which is more applicable to healthcare organizations than most other papers that only predict the number of people involved. However, in terms of the methodology for projecting the number of people, this paper adopts a combined prediction method that combines regression algorithms and recurrent neural networks to propose a BILSTM-GASVR prediction model for the number of confirmed diagnoses. It capitalizes on both the suitability of SVR for small samples and non-linear prediction as well as the learning and memory abilities of BILSTM in processing time-series data. On the basis of the prediction model for the number of infected cases, by considering the characteristics of ICU healthcare resources, we constructed a demand forecasting model of emergency healthcare supplies. Past public health emergencies are more likely to use infectious disease models or a single prediction model in deep learning. some of the articles, although using a combination of prediction, but also more for the same method domain combination, such as CNN-LSTM, GRU-LSTM, etc., which are all recurrent neural networks.

Third, in terms of specific categorization of resources to be forecasted, considering the specificity of ICU medical resources, we introduce human resource prediction on the basis of previous studies focusing on material security, and classified ICU medical resources into three categories: ICU human resources, drugs and medical equipment. The purpose of this classification is to match the real-life prediction scenarios of public health emergencies and improve the demand forecasting performance for local ICU healthcare resources. Thus, it is easy for healthcare institutions to grasp the overall development of events, optimizing decision-making, and reducing the risk of healthcare systems collapsing during the outbreak stage.

In this section, we accomplish the following two tasks. Firstly, we introduce the idea of predicting the number of infected cases and show the principle of the relevant models. Secondly, based on the number of infected cases, ICU healthcare resources are divided into two categories (healthcare workers and healthcare supplies), and their respective demand forecasting models are constructed.

Prediction model for the number of infected cases

Gasvr model.

Support vector machine (SVM) is a machine-learning language for classification developed by Vapnik [ 29 ]. Suppose there are two categories of samples: H1 and H2. If hyperplane H is able to correctly classify the samples into these two categories and maximize the margin between the two categories, it is known as the optimal separating hyperplane (OSH). The sample vectors closest to the OSH in H1 and H2 are known as the support vectors. To apply SVM to prediction, it is essential to perform regression fitting. By introducing the \(\varepsilon\) -insensitive loss function, SVM can be converted to a support vector regression machine, where the role of the OSH is to minimize the error of all samples from this plane. SVR has a theoretical basis in statistical learning and relatively high learning performance, making it suitable for performing predictions in small-sample, non-linear, and multi-dimensional fields [ 30 , 31 ].

Assume the training sample set containing \(l\) training samples is given by \(\{({x}_{i},{y}_{i}),i=\mathrm{1,2},...,l\}\) , where \({x}_{i}=[{x}_{i}^{1},{x}_{i}^{2},...,{x}_{i}^{d}{]}^{\rm T}\) and \({y}_{i}\in R\) are the corresponding output values.

Let the regression function be \(f(x)=w\Phi (x)+b\) , where \(\phi (x)\) is the non-linear mapping function. The linear \(\varepsilon\) -insensitive loss function is defined as shown in formula ( 1 ).

Among the rest, \(f(x)\) is the predicted value returned by the regression function, and \(y\) is the corresponding real value. If the error between \(f(x)\) and \(y\) is ≤ \(\varepsilon\) , the loss is 0; otherwise, the loss is \(\left|y-f(x)\right|-\varepsilon\) .

The slack variables \({\xi }_{i}\) and \({\xi }_{i}^{*}\) are introduced, and \(w\) , \(b\) are solved using the following equation as shown in formula ( 2 ).

Among the rest, \(C\) is the penalty factor, with larger values indicating a greater penalty for errors > \(\varepsilon\) ; \(\varepsilon\) is defined as the error requirement, with smaller values indicating a smaller error of the regression function.

The Lagrange function is introduced to solve the above function and transformed into the dual form to give the formula ( 3 ).

Among the rest, \(K({x}_{i},{x}_{j})=\Phi ({x}_{i})\Phi ({x}_{j})\) is the kernel function. The kernel function determines the structure of high-dimensional feature space and the complexity of the final solution. The Gaussian kernel is selected for this study with the function \(K({x}_{i},{x}_{j})=\mathit{exp}(-\frac{\Vert {x}_{i}-{x}_{j}\Vert }{2{\sigma }^{2}})\) .

Let the optimal solution be \(a=[{a}_{1},{a}_{2},...,{a}_{l}]\) and \({a}^{*}=[{a}_{1}^{*},{a}_{2}^{*},...,{a}_{l}]\) to give the formula ( 4 ) and formula ( 5 ).

Among the rest, \({N}_{nsv}\) is the number of support vectors.

In sum, the regression function is as shown in formula ( 6 ).

when some of the parameters are not 0, the corresponding samples are the support vectors in the problem. This is the principle of SVR. The values of the three unknown parameters (penalty factor C, ε -insensitive loss function, and kernel function coefficient \(\sigma )\) , can directly impact the model effect. The penalty factor C affects the degree of function fitting through the selection of outliers in the sample by the function. Thus, excessively large values lead to better fit but poorer generalization, and vice versa. The ε value in the ε-insensitive loss function determines the accuracy of the model by affecting the width of support vector selection. Thus, excessively large values lead to lower accuracy that does not meet the requirements and excessively small values are overly complex and increase the difficulty. The kernel function coefficient \(\sigma\) determines the distribution and range of the training sample by controlling the size of inner product scaling in high-dimensional space, which can affect overfitting.

Therefore, we introduce other algorithms for optimization of the three parameters in SVR. Currently the commonly used algorithms are 32and some heuristic algorithms. Although the grid search method is able to find the highest classification accuracy, which is the global optimal solution. However, sometimes it can be time-consuming to find the optimal parameters for larger scales. If a heuristic algorithm is used, we could find the global optimal solution without having to trace over all the parameter points in the grid. And GA is one of the most commonly used heuristic algorithms, compared to other heuristic algorithms, it has the advantages of strong global search, generalizability, and broader blending with other algorithms.

Given these factors, we employ a GA to encode and optimize the relevant parameters of the model. The inputs are the experimental training dataset, the Gaussian kernel function expression, the maximum number of generations taken by the GA, the accuracy range of the optimized parameters, the GA population size, the fitness function, the probability of crossover, and the probability of mutation. The outputs are the optimal penalty factor C, ε-insensitive loss function parameter \(\varepsilon ,\) and optimal Gaussian kernel parameter \(\sigma\) of SVR, thus achieving the optimization of SVR. The basic steps involved in GA optimization are described in detail below, and the model prediction process is shown in Fig. 1 .

Prediction process of the GASVR model

Population initialization

The three parameters are encoded using binary arrays composed of 0–1 bit-strings. Each parameter consisted of six bits, and the initial population is randomly generated. The population size is set at 60, and the number of iterations is 200.

Fitness calculation

In the same dataset, the K-fold cross-validation technique is used to test each individual in the population, with K = 5. K-fold cross validation effectively avoids the occurrence of model over-learning and under-learning. For the judgment of the individual, this paper evaluates it in terms of fitness calculations. Therefore, combining the two enables the effective optimization of the model’s selected parameters and improves the accuracy of regression prediction.

Fitness is calculated using the mean error method, with smaller mean errors indicating better fitness. The fitness function is shown in formula ( 7 ) [ 32 ].

The individual’s genotype is decoded and mapped to the corresponding parameter value, which is substituted into the SVR model for training. The parameter optimization range is 0.01 ≤ C ≤ 100, 0.1 ≤ \(\sigma\) ≤ 20, and 0.001 ≤ ε ≤ 1.

Selection: The selection operator is performed using the roulette wheel method.

Crossover: The multi-point crossover operator, in which two chromosomes are selected and multiple crossover points are randomly chosen for swapping, is employed. The crossover probability is set at 0.9.

Mutation: The inversion mutation operator, in which two points are randomly selected and the gene values between them are reinserted to the original position in reverse order, is employed. The mutation probability is set at 0.09.

Decoding: The bit strings are converted to parameter sets.

The parameter settings of the GASVR model built in this paper are shown in Table 1 .

BILSTM model

The LSTM model is a special recurrent neural network algorithm that can remember the long-term dependencies of data series and has an excellent capacity for self-learning and non-linear fitting. LSTM automatically connects hidden layers across time points, such that the output of one time point can arbitrarily enter the output terminal or the hidden layer of the next time point. Therefore, it is suitable for the sample prediction of time-series data and can predict future data based on stored data. Details of the model are shown in Fig. 2 .

Schematic diagram of the LSTM model

LSTM consists of a forget gate, an input gate, and an output gate.

The forget gate combines the previous and current time steps to give the output of the sigmoid activation function. Its role is to screen the information from the previous state and identify useful information that truly impacts the subsequent time step. The equation for the forget gate is shown in formula ( 8 ).

Among the number, \(W_{f}\) is the weight of the forget gate, \({b}_{f}\) is the bias, \(\sigma\) is the sigmoid activation function, \({f}_{t}\) is the output of the sigmoid activation function, \(t-1\) is the previous time step, \(t\) is the current time step, and \({x}_{t}\) is the input time-series data at time step \(t\) .

The input gate is composed of the output of the sigmoid and tanh activation functions, and its role is to control the ratio of input information entering the information of a given time step. The equation for the input gate is shown in formula ( 9 ).

Among the number, \({W}_{i}\) is the output weight of the input gate, \({i}_{t}\) is the output of the sigmoid activation function, \({b}_{i}\) and \({b}_{C}\) are the biases of the input gate, and \({W}_{C}\) is the output of the tanh activation function.

The role of the output gate is to control the amount of information output at the current state, and its equation is shown in formula ( 10 ).

Among the number, \({W}_{o}\) is the weight of \({o}_{t}\) , and \({b}_{o}\) is the bias of the output gate.

The values of the above activation functions \(\sigma\) and tanh are generally shown in formulas ( 11 ) and ( 12 ).

\({C}_{t}\) is the data state of the current time step, and its value is determined by the input information of the current state and the information of the previous state. It is shown in formula ( 13 ).

Among the number, \(\widetilde{{C}_{t}}=\mathit{tan}h({W}_{c}[{h}_{t-1},{x}_{t}]+{b}_{c})\) .

\({h}_{t}\) is the state information of the hidden layer at the current time step, \({h}_{t}={o}_{t}\times \mathit{tan}h({c}_{t})\) .Each time step \({T}_{n}\) has a corresponding state \({C}_{t}\) . By undergoing the training process, the model can learn how to modify state \({C}_{t}\) through the forget, output, and input gates. Therefore, this state is consistently passed on, implying that important distant information will neither be forgotten nor significantly affected by unimportant information.

The above describes the principle of LSTM, which involves forward processing when applied. BILSTM consists of two LSTM networks, one of which processes the input sequence in the forward direction (i.e., the original order), while the other inputs the time series in the backward direction into the LSTM model. After processing both LSTM networks, the outputs are combined, which eventually gives the output results of the BILSTM model. Details of the model are presented in Fig. 3 .

Schematic diagram of the BILSTM model

Compared to LSTM, BILSTM can achieve bidirectional information extraction of the time-series and connect the two LSTM layers onto the same output layer. Therefore, in theory, its predictive performance should be superior to that of LSTM. In BILSTM, the equations of the forward hidden layer( \(\overrightarrow{{h}_{t}}\) ) , backward hidden layer( \(\overleftarrow{{h}_{t}}\) ) , and output layer( \({o}_{t}\) ) are shown in formulas ( 14 ) , ( 15 ) and ( 16 ).

The parameter settings of the BILSTM model built in this paper are shown in Table 2 .

Informer model

The Informer model follows the compiler-interpreter architecture in the Transformer model, and based on this, structural optimizations have been made to reduce the computational time complexity of the algorithm and to optimize the output form of the interpreter. The two optimization methods are described in detail next.

With large amounts of input data, neural network models can have difficulty capturing long-term interdependencies in sequences, which can produce gradient explosions or gradient vanishing and affect the model's prediction accuracy. Informer model solves the existential gradient problem by using a ProbSparse Self-attention mechanism to make more efficient than conventional self-attention.

The value of Transformer self-attention is shown in formula ( 17 ).

Among them, \(Q\in {R}^{{L}_{Q}\times d}\) is the query matrix, \(K\in {R}^{{L}_{K}\times d}\) is the key matrix, and \(V\in {R}^{{L}_{V}\times d}\) is the value matrix, which are obtained by multiplying the input matrix X with the corresponding weight matrices \({W}^{Q}\) , \({W}^{K}\) , \({W}^{V}\) respectively, and d is the dimensionality of Q, K, and V. Let \({q}_{i}\) , \({k}_{i}\) , \(v_{i}\) represent the ith row in the Q, K, V matrices respectively, then the ith attention coefficient is shown in formula ( 18 ) as follows.

Therein, \(p({k}_{j}|{q}_{i})\) denotes the traditional Transformer's probability distribution formula, and \(k({q}_{i},{K}_{l})\) denotes the asymmetric exponential sum function. Firstly, q=1 is assumed, which implies that the value of each moment is equally important; secondly, the difference between the observed distribution and the assumed one is evaluated by the KL scatter, if the value of KL is bigger, the bigger the difference with the assumed distribution, which represents the more important this moment is. Then through inequality \(ln{L}_{k}\le M({q}_{i},K)\le {\mathit{max}}_{j}\left\{\frac{{q}_{i}{k}_{j}^{\rm T}}{\sqrt{d}}\right\}-\frac{1}{{L}_{k}}{\sum }_{j=1}^{{L}_{k}}\left\{\frac{{q}_{i}{k}_{j}^{\rm T}}{\sqrt{d}}\right\}+ln{L}_{k}\) , \(M({q}_{i},K)\) is transformed into \(\overline{M}({q}_{i},K)\) . According to the above steps, the ith sparsity evaluation formula is obtained as shown in formula ( 19 ) [ 33 ].

One of them, \(M({q}_{i},K)\) denotes the ith sparsity measure; \(\overline{M}({q}_{i},K)\) denotes the ith approximate sparsity measure; \({L}_{k}\) is the length of query vector. \(TOP-u\) quantities of \(\overline{M}\) are selected to form \(\overline{Q}\) , \(\overline{Q}\) is the first u sparse matrices, and the final sparse self-attention is shown in Formula ( 20 ). At this point, the time complexity is still \(O({n}^{2})\) , and to solve this problem, only l moments of M2 are computed to reduce the time complexity to \(O(L\cdot \mathit{ln}(L))\) .

Informer uses a generative decoder to obtain long sequence outputs.Informer uses the standard decoder architecture shown in Fig. 4 , in long time prediction, the input given to the decoder is shown in formula ( 21 ).

Informer uses a generative decoder to obtain long sequence outputs

Therein, \({X}_{de}^{t}\) denotes the input to the decoder; \({X}_{token}^{t}\in {R}^{({L}_{token}+{L}_{y})\times {d}_{\mathit{mod}el}}\) is the dimension of the encoder output, which is the starting token without using all the output dimensions; \({X}_{0}^{t}\in {R}^{({L}_{token}+{L}_{y})\times {d}_{\mathit{mod}el}}\) is the dimension of the target sequence, which is uniformly set to 0; and finally the splicing input is performed to the encoder for prediction.

The parameter settings of Informer model created in this paper are shown in Table 3 .

BILSTM-GASVR combined prediction model

SVR has demonstrated good performance in solving problems like finite samples and non-linearity. Compared to deep learning methods, it offers faster predictions and smaller empirical risks. BILSTM has the capacity for long-term memory, can effectively identify data periodicity and trends, and is suitable for the processing of time-series data. Hence, it can be used to identify the effect of time-series on the number of confirmed cases. Given the advantages of these two methods in different scenarios, we combined them to perform predictions using GASVR, followed by error repair using BILSTM. The basic steps for prediction based on the BILSTM-GASVR model are as follows:

Normalization is performed on the initial data.

The GASVR model is applied to perform training and parameter optimization of the data to obtain the predicted value \(\widehat{{y}_{i}}\) .

After outputting the predicted value of GASVR, the residual sequence between the predicted value and real data is extracted to obtain the error \({\gamma }_{i}\) (i.e., \({\gamma }_{i}={y}_{i}-\widehat{{y}_{i}}\) ).

The BILSTM model is applied to perform training of the error to improve prediction accuracy. The BILSTM model in this paper is a multiple input single output model. Its inputs are the true and predicted error values \({\gamma }_{i}\) and its output is the new error value \(\widehat{{\gamma }_{i}}\) predicted by BILSTM.

The final predicted value is the sum of the GASVR predicted value and the BILSTM residual predicted value (i.e., \({Y}_{i}=\widehat{{y}_{i}}+\widehat{{\gamma }_{i}}\) ).

The parameter settings of the BILSTM-GASVR model built in this paper are shown in Table 4 .

Model testing criteria

To test the effect of the model, the prediction results of the BILSTM-GASVR model are compared to those of GASVR, LSTM, BILSTM and Informer. The prediction error is mainly quantified using three indicators: mean squared error (MSE), root mean squared error (RMSE), and correlation coefficient ( \(R^{2}\) ). Their respective equations are shown in formulas ( 22 ), ( 23 ) and ( 24 ).

Demand forecasting model of ICU healthcare resources

ICU healthcare resources can be divided into human and material resources. Human resources refer specifically to the professional healthcare workers in the ICU. Material resources, which are combined with the actual consumption of medical supplies, can be divided into consumables and non-consumables. Consumables refer to the commonly used drugs in the ICU, which include drugs for treating cardiac insufficiency, vasodilators, anti-shock vasoactive drugs, analgesics, sedatives, muscle relaxants, anti-asthmatic drugs, and anticholinergics. Given that public health emergencies have a relatively high probability of affecting the respiratory system, we compiled a list of commonly used drugs for respiratory diseases in the ICU (Table 5 ).

Non-consumables refer to therapeutic medical equipment, including electrocardiogram machines, blood gas analyzers, electrolyte analyzers, bedside diagnostic ultrasound machines, central infusion workstations, non-invasive ventilators, invasive ventilators, airway clearance devices, defibrillators, monitoring devices, cardiopulmonary resuscitation devices, and bedside hemofiltration devices.

The demand forecasting model of ICU healthcare resources constructed in this study, as well as its relevant parameters and definitions, are described below. \({R}_{ij}^{n}\) is the forecasted demand for the \(i\) th category of resources on the \(n\) th day in region \(j\) . \({Y}_{j}^{n}\) is the predicted number of current confirmed cases on the \(n\) th day in region \(j\) . \({M}_{j}^{n}\) is the number of ICU healthcare workers on the \(n\) th day in region \(j\) , which is given by the following formula: number of healthcare workers the previous day + number of new recruits − reduction in number the previous day, where the reduction in number refers to the number of healthcare workers who are unable to work due to infection or overwork. In general, the number of ICU healthcare workers should not exceed 5% of the number of current confirmed cases (i.e., it takes the value range [0, \(Y_{j}^{n}\) ×5%]). \(U_{i}\) is the maximum working hours or duration of action of the \(i\) th resource category within one day. \({A}_{j}\) is the number of resources in the \(i\) th category allocated to patients (i.e., how many units of resources in the \(i\) th category is needed for a patient who need the \(i\) th unit of the given resource). \({\varphi }_{i}\) is the demand conversion coefficient (i.e., the proportion of the current number of confirmed cases who need to use the \(i\) th resource category). \({C}_{ij}^{n}\) is the available quantity of material resources of the \(i\) th category on the \(n\) th day in region \(j\) . At the start, this quantity is the initial reserve, and once the initial reserve is exhausted, it is the surplus from the previous day. The formula for this parameter is given as follows: available quantity from the previous day + replenishment on the previous day − quantity consumed on the previous day, where if \({C}_{ij}^{n}\) is a negative number, it indicates the amount of shortage for the given category of resources on the previous day.

In summary, the demand forecast for emergency medical supplies constructed in this study is shown in formula ( 25 ).

The number of confirmed cases based on data-driven prediction is introduced into the demand forecasting model for ICU resources to forecast the demand for the various categories of resources. In addition to the number of current confirmed cases, the main variables of the first demand forecasting model for human resources are the available quantity and maximum working hours. The main variable of the second demand forecasting model for consumable resources is the number of units consumed by the available quantity. The main variable of the third model for non-consumable resources is the allocated quantity. These three resource types can be predicted using the demand forecasting model constructed in this study.

Prediction of the number of current infected cases

The COVID-19 situation in Shanghai is selected for our experiment. A total of 978 entries of epidemic-related data in Shanghai between January 20, 2020, and September 24, 2022, are collected from the epidemic reporting platform. This dataset is distributed over a large range and belongs to a right-skewed leptokurtic distribution. The specific statistical description of data is shown in Table 6 . Part of the data is shown in Table 7 .

And we divided the data training set and test set in an approximate 8:2 ratio, namely, 798 days for training (January 20, 2020 to March 27, 2022) and 180 days for prediction (March 28, 2022 to September 24, 2022).

Due to the large difference in order of magnitude between the various input features, directly implementing training and model construction would lead to suboptimal model performance. Such effects are usually eliminated through normalization. In terms of interval selection, [0, 1] reflects the probability distribution of the sample, whereas [-1, 1] mostly reflects the state distribution or coordinate distribution of the sample. Therefore, [-1, 1] is selected for the normalization interval in this study, and the processing method is shown in formula ( 26 ).

Among the rest, \(X\) is the input sample, \({X}_{min}\) and \({X}_{max}\) are the minimum and maximum values of the input sample, and \({X}_{new}\) is the input feature after normalization.

In addition, we divide the data normalization into two parts, considering that the amount of data in the training set is much more than the test set in the real operating environment. In the first step, we normalize the training set data directly according to the above formula; in the second step, we normalize the test data set using the maximum and minimum values of the training data set.

The values of the preprocessed data are inserted into the GASVR, LSTM, Informer, BILSTM models and the BILSTM-GASVR model is constructed. Figures 5 , 6 , 7 , 8 and 9 show the prediction results. From Figs. 5 , 6 , and 7 , it can be seen that in terms of data accuracy, GASVR more closely matches the real number of infected people relative to BILSTM and LSTM. Especially in the most serious period of the epidemic in Shanghai (April 17, 2022 to April 30, 2022), the advantage of the accuracy of the predicted data of GASVR is even more obvious, which is due to the characteristics of GASVR for small samples and nonlinear prediction. However, in the overall trend of the epidemic, BILSTM and LSTM, which have the ability to learn and memorize to process time series data, are superior. It is clearly seen that in April 1, 2022-April 7, 2022 and May 10, 2022-May 15, 2022, there is a sudden and substantial increase in GASVR in these two time phases, and a sudden and substantial decrease in April 10, 2022-April 14, 2022. These errors also emphasize the stability of BILSTM and LSTM, which are more closely matched to the real epidemic development situation in the whole process of prediction, and the difference between BILSTM and LSTM prediction is that the former predicts data more accurately than the latter, which is focused on the early stage of prediction as well as the peak period of the epidemic. Informer is currently an advanced time series forecasting method. From Fig. 8 , it can be seen that the prediction data accuracy and the overall trend of the epidemic are better than the single prediction models of GASVR, LSTM and BILSTM. However, Informer is more suitable for long time series and more complex and large prediction problems, so the total sample size of less than one thousand cases is not in the comfort zone of Informer model. Figure 9 shows that the BILSTM-GASVR model constructed in this paper is more suitable for this smaller scale prediction problem, with the best prediction results, closest to the actual parameter (number of current confirmed cases), demonstrating small sample and time series advantages. In Short, the prediction effect of models is ranked as follows: BILSTM-GASVR> Informer> GASVR> BILSTM> LSTM.

The prediction result of the GASVR model

The prediction result of the LSTM model

The prediction result of the BILSTM model

The prediction result of the Informer model

The prediction result of the BILSTM-GASVR model

The values of the three indicators (MSE, RMSE, and correlation coefficient \({R}^{2}\) ) for the five models are shown in Table 8 . MSE squares the error so that the larger the model error, the larger the value, which help capture the model's prediction error more sensitively. RMSE is MSE with a root sign added to it, which allows for a more intuitive representation of the order of magnitude difference from the true value. \({R}^{2}\) is a statistical indicator used to assess the overall goodness of fit of the model, which reflects the overall consistency of the predicted trend and does not specifically reflect the degree of data. The results in the Table 8 are consistent with the prediction results in the figure above, while the ranking of MSE, RMSE, and \({R}^{2}\) are also the same (i.e., BILSTM-GASVR> Informer> GASVR> BILSTM> LSTM).

In addition, we analyze the five model prediction data using significance tests as a way of demonstrating whether the model used is truly superior to the other baseline models. The test dataset with kurtosis higher than 4 does not belong to the approximate normal distribution, so parametric tests are not used in this paper. Given that the datasets predicted by each of the five models are continuous and independent datasets, this paper uses the Kruskal-Wallis test, which is a nonparametric test. The test steps are as follows.

Determine hypotheses (H0, H1) and significance level ( \(\alpha\) ).

For each data set, all its sample data are combined and ranked from smallest to largest. Then find the number of data items ( \({n}_{i}\) ), rank sum ( \({R}_{i}\) ) and mean rank of each group of data respectively.

Based on the rank sum, the test statistic (H) is calculated for each data set in the Kruskal-Wallis test. The specific calculation is shown in formula ( 27 ).

According to the test statistic and degrees of freedom, find the corresponding p-value in the Kruskal-Wallis distribution table. Based on the P-value, determine whether the original hypothesis is valid.

In the significance test, we set the significance setting original hypothesis (H0) as there is no significant difference between the five data sets obtained from the five predictive models. We set the alternative hypothesis (H1) as there is a significant difference between the five data sets obtained from the five predictive models. At the same time, we choose the most commonly used significance level taken in the significance test, namely 0.05. In this paper, multiple comparisons and two-by-two comparisons of the five data sets obtained from the five predictive models are performed through the SPSS software. The results of the test show that in the multiple comparison session, P=0.001<0.05, so H0 is rejected, which means that the difference between the five groups of data is significant. In the two-by-two comparison session, BILSTM-GASVR is less than 0.05 from the other four prediction models. The specific order of differences is Informer < GASVR < BILSTM < LSTM, which means that the BILSTM-GASVR prediction model does get a statistically significant difference between the dataset and the other models.

In summary, combined prediction using the BILSTM-GASVR model is superior to the other four single models in various aspects in the case study analysis of Shanghai epidemic with a sample size of 978.

Demand forecasting of ICU healthcare resources

Combined with the predicted number of current infected cases, representatives are selected from the three categories of resources for forecasting. The demand for nurses is selected as the representative for the first category of resources.

In view of the fact that there are currently no specific medications that are especially effective for this public health emergency, many ICU treatment measures involved helping patients survive as their own immune systems eliminated the virus. This involved, for example, administering antibiotics when patients developed a secondary bacterial infection. glucocorticoids are used to temporarily suppress the immune system when their immune system attacked and damaged lung tissues causing patients to have difficulty breathing. extracorporeal membrane oxygenation (ECMO) is used for performing cardiopulmonary resuscitation when patients are suffering from cardiac arrest. In this study, we take dexamethasone injection (5 mg), a typical glucocorticoid drug, as the second category of ICU resources (i.e., drugs); and invasive ventilators as the third category of ICU resources (i.e., medical equipment).

During the actual epidemic in Shanghai, the municipal government organized nine critical care teams, which are stationed in eight municipally designated hospitals and are dedicated to the treatment of critically ill patients. In this study, the ICU nurses, dexamethasone injections, and invasive ventilators in Shanghai are selected as the prediction targets and introduced into their respective demand forecasting models. Forecasting of ICU healthcare resources is then performed for the period from March 28, 2022, to April 28, 2022, as an example. Part of the parameter settings for the three types of resources are shown in Tables 9 , 10 , and 11 , respectively.

Table 12 shows the forecasting results of the demand for ICU nurses, dexamethasone injections, and invasive ventilators during the epidemic wave in Shanghai between March 28, 2022, and April 28, 2022.

For the first category (i.e., ICU nurses), human resource support is only needed near the peak period, but the supply could not be replenished immediately. In the early stages, Shanghai could only rely on the nurses’ perseverance, alleviating the shortage of human resources by reducing the number of shifts and increasing working hours. This situation persisted until about April 10 and is only resolved when nurses from other provinces and regions successively arrived in Shanghai.

The second category of ICU resources is drugs, which are rapidly consumed. The pre-event reserve of 30,000 dexamethasone injections could only be maintained for a short period and is fully consumed during the outbreak. Furthermore, daily replenishment is still needed, even when the epidemic has passed its peak and begun its decline.

The third category is invasive ventilators, which are non-consumables. Thus, the reserve lasted for a relatively long period of time in the early stages and did not require replenishment after its maximum usage during the peak period.

Demand forecasting models are constructed based on the classification of healthcare resources according to their respective features. We choose ICU nurses, dexamethasone injections, and invasive ventilators as examples, and then forecast demand for the epidemic wave in Shanghai between March 28, 2022, and April 28, 2022. The main conclusions are as follows:

A long period of time is needed to train ICU healthcare workers who can independently be on duty, taking at least one year from graduation to entering the hospital, in addition to their requiring continuous learning, regular theoretical training, and the accumulation of clinical experience during this process. Therefore, for the first category of ICU healthcare resources, in the long term, healthcare institutions should place a greater emphasis on their talent reserves. Using China as an example, according to the third ICU census, the ratio of the number of ICU physicians to the number of beds is 0.62:1 and the ratio of the number of nurses to the number of beds is 1.96:1, which are far lower than those stipulated by China itself and those of developed countries. Therefore, a fundamental solution is to undertake proactive and systematic planning and construction to ensure the more effective deployment of human resources in the event of a severe outbreak. In the short term, healthcare institutions should focus on the emergency expansion capacity of their human resources. In case there are healthcare worker shortages during emergencies, the situation can be alleviated by summoning retired workers back to work and asking senior medical students from various universities to help in the hospitals to prevent the passive scenario of severely compressing the rest time of existing staff or waiting for external aid. However, it is worth noting that to ensure the effectiveness of such a strategy of using retired healthcare workers or senior students of university medical faculties, it is necessary for healthcare organizations to provide them with regular training in the norm, such as organizing 2-3 drills a year, to ensure the professionalism and proficiency of healthcare workers who are temporarily and suddenly put on the job. At the same time, it is also necessary to fully mobilize the will of individuals. Medical institutions can provide certain subsidies to retired health-care workers and award them with honorable titles. For senior university medical students, volunteer certificates are issued and priority is given to their internships, so that health-care workers can be motivated to self-realization through spiritual and material rewards.

Regarding the second category of ICU resources (i.e., drugs), healthcare institutions perform the subdivision of drug types and carry out dynamic physical preparations based on 15–20% of the service recipient population for clinically essential drugs. This will enable a combination of good preparedness during normal times and emergency situations. In addition, in-depth collaboration with corporations is needed to fully capitalize on their production capacity reserves. This helps medical institutions to be able to scientifically and rationally optimize the structure and quantity of their drug stockpiles to prevent themselves from being over-stressed. Yet the lower demand for medicines at the end of the epidemic led to the problem of excess inventory of enterprises at a certain point in time must be taken into account. So, the medical institutions should sign a strategic agreement on stockpiling with enterprises, take the initiative to bear the guaranteed acquisition measures, and consider the production costs of the cooperative enterprises. These measures are used to truly safeguard the enthusiasm of the cooperative enterprises to invest in the production capacity.

Regarding the third category of ICU resources (i.e., medical equipment), large-scale medical equipment cannot be rapidly mass-produced due to limitations in the capacity for emergency production and conversion of materials. In addition, the bulk procurement of high-end medical equipment is also relatively difficult in the short term. Therefore, it is more feasible for healthcare institutions to have physical reserves of medical equipment, such as invasive ventilators. However, the investment costs of medical equipment are relatively high. Ventilators, for example, cost up to USD $50,000, and subsequent maintenance costs are also relatively high. After all, according to the depreciable life of specialized hospital equipment, the ventilator, as a surgical emergency equipment, is depreciated over five years. And its depreciation rate is calculated at 20% annually for the first five years, which means a monthly depreciation of $835. Thus, the excessively low utilization rate of such equipment will also impact the hospital. Healthcare institutions should, therefore, conduct further investigations on the number of beds and the reserves of ancillary large-scale medical equipment to find a balance between capital investment and patient needs.

The limitations of this paper are reflected in the following three points. Firstly, in the prediction of the number of infections, the specific research object in this paper is COVID-19, and other public health events such as SARS, H1N1, and Ebola are not comparatively analyzed. The main reason for this is the issue of data accessibility, and it is easier for us to analyze events that have occurred in recent years. In addition, using the Shanghai epidemic as a specific case may be more representative of the epidemic situation in an international metropolis with high population density and mobility. Hence, it has certain regional limitations, and subsequent studies should expand the scope of the case study to reflect the characteristics of epidemic transmission in different types of urban areas and enhance the generalizability.

Secondly, the main emphasis of this study is on forecasting the demand for ICU healthcare resources across the entire region of the epidemic, with a greater focus on patient demand during public emergencies. Our aims are to help all local healthcare institutions more accurately identify changes in ICU healthcare resource demand during this local epidemic wave, gain a more accurate understanding of the treatment demands of critically ill patients, and carry out comprehensive, scientifically based decision-making. Therefore, future studies can examine individual healthcare institutions instead and incorporate the actual conditions of individual units to construct multi-objective models. In this way, medical institutions can further grasp the relationship between different resource inputs and the recovery rate of critically ill patients, and achieve the balance between economic and social benefits.

Finally, for the BILSTM-GASVR prediction method, in addition to the number of confirmed diagnoses predicted for an outbreak in a given region, other potential applications beyond this type of medium-sized dataset still require further experimentation. For example, whether the method is suitable for procurement planning of a certain supply in production management, forecasting of goods sales volume in marketing management, and other long-period, large-scale and other situations.

Within the context of major public health events, the fluctuations and uncertainties in the demand for ICU resources can lead to large errors between the healthcare supply and actual demand. Therefore, this study focuses on the question of forecasting the demand for ICU healthcare resources. Based on the number of current confirmed cases, we construct the BILSTM-GASVR model for predicting the number of patients. By comparing the three indicators (MSE, MAPE, and correlation coefficient \(R^{2}\) ) and the results of the BILSTM, LSTM, and GASVR models, we demonstrate that our model have a higher accuracy. Our findings can improve the timeliness and accuracy of predicting ICU healthcare resources and enhance the dynamics of demand forecasting. Hence, this study may serve as a reference for the scientific deployment of ICU resources in healthcare institutions during major public events.

Given the difficulty in data acquisition, only the Shanghai epidemic dataset is selected in this paper, which is one of the limitations mentioned in Part 4. Although the current experimental cases of papers in the same field do not fully conform to this paper, the results of the study cannot be directly compared. However, after studying the relevant reviews and the results of the latest papers, we realize that there is consistency in the prediction ideas and prediction methods [ 34 , 35 ]. Therefore, we summarize the similarities and differences between the results of the study and other research papers in epidemic forecasting as shown below.

Similarities: on the one hand, we all characterize trends in the spread of the epidemic and predict the number of infections over 14 days. On the other hand, we all select the current mainstream predictive models as the basis and combine or improve them. Moreover, we all use the same evaluation method (comparison of metrics such as MSE and realistic values) to evaluate the improvements against other popular predictive models.

Differences: on the one hand, other papers focus more on predictions at the point of the number of patients, such as hospitalization rate, number of infections, etc. This paper extends the prediction from the number of patients to the specific healthcare resources. This paper extends the prediction from the number of patients to specific healthcare resources. We have divided the medical resources and summarized the demand regularities of the three types of information in the epidemic, which provides the basis for decision-making on epidemic prevention to the government or medical institutions. On the other hand, in addition to the two assessment methods mentioned in the same point, this paper assesses the performance of the prediction methods with the help of significance tests, which is a statistical approach to data. This can make the practicality of the forecasting methodology more convincing.

Availability of data and materials

The datasets used and/or analysed during the current study available from the corresponding author on reasonable request.

Yuan, R., Yang, Y., Wang, X., Duo, J. & Li, J. Study on the forecasting and allocation of emergency medical material needs in the event of a major infectious disease outbreak. J Saf Environ. https://doi.org/10.13637/j.issn.1009-6094.2023.2448 .

Total epidemic data (global). 2024. Retrieved March 30, 2024, from: https://www.sy72.com/world/ .

Sui, K., Wang, Y., Wang, S., Chen, C., & Sun, X. A multiple regression analysis-based method for forecasting the demand of emergency materials for power grids. Electron Technol Softw Eng. 2016; (23), 195-197.

Li K, Liu L, Zhai J, Khoshgoftaar TM, Li T. The improved grey model based on particle swarm optimization algorithm for time series prediction. Eng Appl Artif Intell. 2016;55:285–91. https://doi.org/10.1016/j.engappai.2016.07.005 .

Article   Google Scholar  

Fan R, Wang Y, Luo M. SEIR-based COVID-19 transmission model and inflection point prediction analysis. J Univ Electron Sci Technol China. 2020;49(3):369–74.

Google Scholar  

Neves AGM, Guerrero G. Predicting the evolution of the COVID-19 epidemic with the A-SIR model: Lombardy, Italy and São Paulo state Brazil. Physica D. 2020;413:132693. https://doi.org/10.1016/j.physd.2020.132693 .

Article   PubMed   PubMed Central   Google Scholar  

Li R, Pei S, Chen B, Song Y, Zhang T, Wan Y, Shaman J. Substantial undocumented infection facilitates the rapid dissemination of novel coronavirus (SARS-CoV-2). Science. 2020;368(6490):489–93. https://doi.org/10.1126/science.abb3221 .

Article   CAS   PubMed   PubMed Central   Google Scholar  

Kermack WO, McKendrick AG. A contribution to the mathematical theory of epidemics. Proc R Soc Lond. 1927;115(772):700–21. https://doi.org/10.1098/rspa.1927.0118 .

Hethcote HW. The mathematics of infectious diseases. Siam Rev. 2000;42(4):599–653. https://doi.org/10.1137/SIREAD000042000004000655000001 .

Estrada E. COVID-19 and SARS-CoV-2. Modeling the present, looking at the future. Phys Rep. 2020;869:1–51. https://doi.org/10.1016/j.physrep.2020.07.005 .

Mizumoto K, Chowell G. Transmission potential of the novel coronavirus (COVID-19) onboard the diamond Princess Cruises Ship, 2020. Infect Dis Model. 2020;5:264–70. https://doi.org/10.1016/j.idm.2020.02.003 .

Delamater PL, Street EJ, Leslie TF, Yang YT, Jacobsen KH. Complexity of the Basic Reproduction Number. Emerg Infect Dis. 2019;25(1):1–4. https://doi.org/10.3201/eid2501.171901 .

Annas S, Isbar Pratama M, Rifandi M, Sanusi W, Side S. Stability analysis and numerical simulation of SEIR model for pandemic COVID-19 spread in Indonesia. Chaos Solit Fract. 2020;139:110072. https://doi.org/10.1016/j.chaos.2020.110072 .

Li Q, Guan X, Wu P, Wang X, Zhou L, Tong Y, Ren R, Leung KSM. Early Transmission Dynamics in Wuhan, China, of Novel Coronavirus-Infected Pneumonia. N Engl J Med. 2020;382(13):1199–209. https://doi.org/10.1056/NEJMoa2001316 .

Anggriani N, Ndii MZ, Amelia R, Suryaningrat W, Pratama MAA. A mathematical COVID-19 model considering asymptomatic and symptomatic classes with waning immunity. Alexandria Eng J. 2022;61(1):113–24. https://doi.org/10.1016/j.aej.2021.04.104 .

Efimov D, Ushirobira R. On an interval prediction of COVID-19 development based on a SEIR epidemic model. Ann Rev Control. 2021;51:477–87. https://doi.org/10.1016/j.arcontrol.2021.01.006 .

Lin Q, Zhao S, Gao D, Lou Y, Yang S, Musa SS, Wang MH. A conceptual model for the coronavirus disease 2019 (COVID-19) outbreak in Wuhan, China with individual reaction and governmental action. Int J Infect Dis. 2020;93:211–16.

Chao L, Feng P, Shi P. Study on the epidemic development of COVID-19 in Hubei. J Zhejiang Univ (Med Sci). 2020;49(2):178–84.

Reiner RC, Barber RM, Collins JK, et al. Modeling COVID-19 scenarios for the United States. Nat Med. 2021;27(1):94–105. https://doi.org/10.1038/s41591-020-1132-9 .

Article   CAS   Google Scholar  

Shahid F, Zameer A, Muneeb M. Predictions for COVID-19 with deep learning models of LSTM GRU and Bi-LSTM. Chaos Solit Fract. 2020;140:110212. https://doi.org/10.1016/j.chaos.2020.110212 .

Chimmula VKR, Zhang L. Time series forecasting of COVID-19 transmission in Canada using LSTM networks. Chaos Solit Fract. 2020;135:109864. https://doi.org/10.1016/j.chaos.2020.109864 .

Hamou AA, Azroul E, Hammouch Z, Alaoui AAL. A fractional multi-order model to predict the COVID-19 outbreak in Morocco. Appl Comput Math. 2021;20(1):177–203.

Zhou, L., Zhao, C., Liu, N., Yao, X., & Cheng, Z. Improved LSTM-based deep learning model for COVID-19 prediction using optimized approach. Eng Appl Artif Intell. 2023; 122. https://doi.org/10.1016/j.engappai.2023.106157 .

Huang C, Chen Y, Ma Y, Kuo P. Multiple-Input Deep Convolutional Neural Network Model for COVID-19 Forecasting in China. medRxiv. 2020;74822–34.

Gautam Y. Transfer Learning for COVID-19 cases and deaths forecast using LSTM network. Isa Trans. 2022;124:41–56. https://doi.org/10.1016/j.isatra.2020.12.057 .

Article   PubMed   Google Scholar  

Ghany KKA, Zawbaa HM, Sabri HM. COVID-19 prediction using LSTM algorithm: GCC case study. Inform Med Unlock. 2021;23:100566. https://doi.org/10.1016/j.imu.2021.100566 .

Devaraj J, Madurai Elavarasan R, Pugazhendhi R, Shafiullah GM, Ganesan S, Jeysree AK, Khan IA, Hossain E. Forecasting of COVID-19 cases using deep learning models: Is it reliable and practically significant? Results Phys. 2021;21:103817. https://doi.org/10.1016/j.rinp.2021.103817 .

Arora P, Kumar H, Panigrahi BK. Prediction and analysis of COVID-19 positive cases using deep learning models: a descriptive case study of India. Chaos Solit Fract. 2020;139:110017. https://doi.org/10.1016/j.chaos.2020.110017 .

Liu Q, Fung DLX, Lac L, Hu P. A Novel Matrix Profile-Guided Attention LSTM Model for Forecasting COVID-19 Cases in USA. Front Public Health. 2021;9:741030. https://doi.org/10.3389/fpubh.2021.741030 .

Ribeiro MHDM, Da Silva RG, Mariani VC, Coelho LDS. Short-term forecasting COVID-19 cumulative confirmed cases: perspectives for Brazil. Chaos Solit Fract. 2020;135:109853. https://doi.org/10.1016/j.chaos.2020.109853 .

Shoko C, Sigauke C. Short-term forecasting of COVID-19 using support vector regression: an application using Zimbabwean data. Am J Infect Control. 2023. https://doi.org/10.1016/j.ajic.2023.03.010 .

Feng T, Peng Y, Wang J. ISGS: A Combinatorial Model for Hysteresis Effects. Acta Electronica Sinica. 2023;09:2504–9.

Zhou H, Zhang S, Peng J, Zhang S, Li J, Xiong H, Zhang W. Informer: Beyond Efficient Transformer for Long Sequence Time-Series Forecasting. AAAI Conference Artif Intell. 2020;35(12):11106–15.

Rahimi I, Chen F, Gandomi AH. A review on COVID-19 forecasting models. Neural Comput Applic. 2023;35:23671–81. https://doi.org/10.1007/s00521-020-05626-8 .

Chen J, Creamer GG, Ning Y, Ben-Zvi T. Healthcare Sustainability: Hospitalization Rate Forecasting with Transfer Learning and Location-Aware News Analysis. Sustainability. 2023;15(22):15840. https://doi.org/10.3390/su152215840 .

Download references

Acknowledgements

We would like to acknowledge the hard and dedicated work of all the staff that implemented the intervention and evaluation components of the study.

No external funding received to conduct this study.

Author information

Authors and affiliations.

School of Logistics, Beijing Wuzi University, No.321, Fuhe Street, Tongzhou District, Beijing, 101149, China

Weiwei Zhang & Xinchun Li

You can also search for this author in PubMed   Google Scholar

Contributions

WWZ and XCL conceived the idea and conceptualised the study. XCL collected the data. WWZ analysed the data. WWZ and XCL drafted the manuscript, then WWZ and XCLreviewed the manuscript. WWZ and XCL read and approved the final draft.

Corresponding author

Correspondence to Xinchun Li .

Ethics declarations

Ethics approval and consent to participate.

Not applicable.

Consent for publication

Competing interests.

The authors declare no competing interests.

Additional information

Publisher’s note.

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/ . The Creative Commons Public Domain Dedication waiver ( http://creativecommons.org/publicdomain/zero/1.0/ ) applies to the data made available in this article, unless otherwise stated in a credit line to the data.

Reprints and permissions

About this article

Cite this article.

Zhang, W., Li, X. A data-driven combined prediction method for the demand for intensive care unit healthcare resources in public health emergencies. BMC Health Serv Res 24 , 477 (2024). https://doi.org/10.1186/s12913-024-10955-8

Download citation

Received : 21 September 2023

Accepted : 05 April 2024

Published : 17 April 2024

DOI : https://doi.org/10.1186/s12913-024-10955-8

Share this article

Anyone you share the following link with will be able to read this content:

Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative

• Public health emergency
• ICU healthcare resource demand
• Machine learning
• Combined prediction

BMC Health Services Research

ISSN: 1472-6963

• General enquiries: [email protected]