Coronavirus Disease Outbreak Prediction and Analysis Using Machine Learning and Classical Time Series Forecasting Models

This study explores the efficacy of machine learning and classical time series forecasting models in predicting COVID-19 case counts, addressing a critical global health challenge. We compared four models—Holt-Winters, ARIMA, Facebook Prophet, and XGBoost—across eight countries, forecasting both confirmed and death cases over a 14-day period. Our findings reveal that Facebook Prophet with holiday impact and XGBoost consistently outperform other models in predicting new cases, with XGBoost excelling in forecasting death cases, followed by ARIMA and Facebook Prophet. Notably, XGBoost with a mean absolute error (MAE) of 1.42, mean squared error (MSE) of 4.06, and root mean squared error (RMSE) of 2.01, demonstrated the best performance for death prediction. These results underscore the potential of advanced analytical techniques in pandemic management, offering valuable insights for public health officials and policymakers. By providing accurate short-term projections, these models can facilitate proactive decision-making, resource allocation, and intervention strategies, ultimately contributing to more effective COVID-19 mitigation efforts worldwide. This re- search not only demonstrates the practical application of machine learning in epidemiological forecasting but also highlights the importance of model selection in achieving optimal predictive performance for different aspects of the pandemic.