Forecasting Techniques for the Pandemic Trend of COVID-19

Forecasting Techniques for the Pandemic Trend of COVID-19

Copyright: © 2022 |Pages: 27
DOI: 10.4018/978-1-7998-8793-5.ch003
OnDemand:
(Individual Chapters)
Available
$37.50
No Current Special Offers
TOTAL SAVINGS: $37.50

Abstract

Forecasting the trend of the COVID-19 pandemic has been crucial for controlling the spread and making related disease control policies. Various forecasting techniques can be served thereby assisting in strengthening the healthcare system to fight the pandemic. With the development of big data and machine learning techniques, prediction models become more accurate in yielding preparations against risks and threats. In this chapter, three types of forecasting methods, machine learning models, time series forecasting techniques, and deep learning algorithms, are categorized and introduced, mathematically and empirically. To justify the outcomes from each model, this chapter has presented case studies of three pandemic scenarios, including the early stage, the second wave, and the real-time prediction, with real data for the United States. Model comparisons and evaluations have been also illustrated to forecast the number of possible causes. Various existing studies about pandemic predictions are included in the current research by big data analytics.
Chapter Preview
Top

Introduction

Over the last decade, data-driven approaches have proved to be problem-solving kernels for many complicated and sophisticated real-world applications. Such methods are relied on statistical inferences, data analysis processes, machine learning algorithms, and deep learning architectures, which follow the programming reinforcements based on decision-making requiems, such as predictive analysis, data mining, and what-if analysis. The forecasting technique, as one of the most powerful and critical research domains in data-driven decision science, has been widely employed to guide the future course of disease projection and prognosis. Various statistical modeling methods and classical machine learning algorithms have been applied in medical and epidemiological areas to predict the future conditions of a specific disease, along with big data analytics for the healthcare system (Chen et al., 2017). With the rapid rise of deep learning during the past few years, a variety of studies have been performed for disease predictions, disease risk factors, and disease-genetic associations (Zhou et al., 2018; Luo et al., 2019; Ali et al., 2020). Particularly, most recent studies are focused on the real-time forecasting of the spreading features for novel coronavirus and the prediction of Covid-19 outbreak and its early response (Grasselli et al., 2020). Such a prediction system can be used in constructing a smart healthcare framework that handles the current situation by instructing the public authorities to make early prevention measures and interventions to control the new disease effectively.

With the number of Covid-19 cases has been increasing rapidly, the public authorities and individuals keep eyes on the peak and the duration of Covid-19. Policy-makers are mostly concerned with the number of infected cases and deaths in the next few weeks. Although the pandemic will eventually end, key questions of the public healthcare system remain. Can the public health system issue a pandemic alarm based on daily updated information? The question can be answered by forecasting the trend of Covid-19 using time series models based on the daily reported data. In the early stage of the outbreak, no historical data is available to guide model building, however, as the pandemic spreads and volumes of data are amassed, new insights from the prediction models have been captured. Forecasting techniques are critical in terms of better understanding the current health situation and projections of the pandemic trend. Applying appropriate models and increasing their accuracy consistently can enhance the decision-making process of the public health in order to prepare for answering the shock from this ongoing global pandemic. From the computational perspectives, fitting a forecasting model only takes a few seconds, which can be easily applied to the research studies and industrial applications. Data used for the model building relies on daily reports that can be reached from many data sources. The accuracy of the forecasting depends on updating the models in a daily manner for model development, parameter setting, and estimating prediction intervals (Petropoulos & Makridakis, 2020).

Visual mining and forecasting of the number of Covid-19 cases have become the major component for promoting the application of the computational power and resource in smart healthcare management. Several curve-fitting techniques have been used to build the predictive model that serves as the forecasting engine in the dashboards, such as the confirmed and forecasted case data model introduced by the Los Alamos National Laboratory (Los Alamos, 2020), an interactive real-time Covid-19 cases tracker called CovidCounties (Arneson et al., 2020), the worldwide coronavirus outbreak data analysis and prediction system named as CoronaTracker (Hamzah et al., 2020). However, their forecasting results vary over time and regions as they deployed different predictive models with various kinds of input variables and parameters. As the predictive results from such studies/applications are far away from the actual values, it is important to check which models are the most effective for forecasting the Covid-19 cases. Moreover, the model fitting and predicting time are also required to be taken into account in terms of the model efficiency and complexity.

Key Terms in this Chapter

Machine Learning: A subject of artificial intelligence that aims at the task of computational algorithms, which allow machines to learning objects automatically through historical data.

LSTM: An artificial recurrent neural network architecture that has advantages of processing with sequential data.

SVM: A supervised learning method that analyze data for classification and regression problems by creating a line or a hyperplane which separate the data into classes.

Deep Learning: A subset of machine learning based on artificial neural networks with more layers, which can improve the model performance significantly.

Time-Series Forecasting: A prediction method for time-based historical data to build models to forecast future based on data characteristics.

Early Disease Prevention: A series of measures that prevent an upcoming illness from widely spreading.

ARIMA: A model that is used for time series analysis to characterize changes over time through statistical methods.

Curve-Fitting Technique: A method of constructing a mathematical function and adjusting it in order to achieve the best fit to all data points.

Complete Chapter List

Search this Book:
Reset