Comparative Study of Machine Learning Models for Predicting COVID-19

Amid the ongoing global COVID-19 pandemic's significant impact on societies since 2019, accurate predictions of its future spread have gained urgency. This study aims to evaluate four models—two-time series (ARIMA and Prophet) and two Machine Learning (Random Forest and Light Gradient Boosting Machine)—for predicting COVID-19 case trends in West African countries. Utilizing COVID-19 case data from Johns Hopkins University, the study employs these models in repeated experiments to forecast future trends, evaluating their performance using mean absolute error (MAE). The ARIMA model consistently outperforms the others, displaying notably low MAE values across all countries. The Prophet model follows with the second-best performance, particularly effective for certain nations, while the remaining models demonstrate weaker predictive abilities. The research underscores ARIMA's exceptional accuracy in forecasting COVID-19 transmission, suggesting its potential to bolster preparedness and response strategies in West African countries for controlling the virus's spread, despite limitations inherent in model predictions.