(BigData 2024) How Effective are Time Series Models for Pandemic Forecasting?

The increasing frequency and impact of pandemics highlight the need for better preparedness and accurate prediction of infectious disease outbreaks. While traditional tools remain valuable, the focus on predicting infectious diseases has shifted towards time series analysis using statistical and machine learning models. The emergence of a simple linear model has shaken the dominance of complex Transformer-based models in time series forecasting. This unexpected challenge compels us to reevaluate the time series forecasting models, emphasizing the need to distinguish effectiveness from complexity. This research analyzes the performance of sixteen time series models, including simple and complex architectures, in forecasting pandemic data. We evaluate the models on the United States weekly influenza-like illness cases and COVID-19 incident deaths. By evaluating models on datasets of different scales, we aim to identify effective forecasting methods for early-stage and later-stage pandemics. In line with our broader reevaluation of time series models, we also examine the accuracy of the Centers for Disease Control and Prevention models predicting COVID-19 weekly incident deaths on a national level.