Leveraging Artificial Intelligence in Public Health: A Comparative Evaluation of Machine-Learning Algorithms in Predicting COVID-19 Mortality

Background: The COVID-19 pandemic led to millions of deaths worldwide, underscoring the critical role of advanced data-driven approaches in public health crisis management. Machine learning (ML) methods have gained prominence in predicting disease trends, yet comparative analyses of their effectiveness in forecasting COVID-19 mortality remain limited. Objective: This study aimed to evaluate and compare the predictive performance of four ML algorithms – K-Nearest Neighbors (KNN), Random Forest, Extreme Gradient Boosting (XGBoost), and Decision Tree – in estimating daily new COVID-19 deaths. Methods: Using worldwide data from the Our World in Data COVID-19 dataset, daily mortality from March 2020 to May 2023 was modeled using 16 sociodemographic, healthcare, and policy-related predictor variables. The four ML models were trained and tested on this dataset, with performance assessed using R2, Mean Absolute Error (MAE), and Root Mean Squared Error (RMSE). Results: Random Forest achieved the best predictive performance (R2 = 0.983, MAE = 0.61, RMSE = 2.79), followed closely by XGBoost. Decision Tree and KNN exhibited weaker predictive accuracy with lower R2 values and higher error metrics. Across the four models, key predictors of COVID-19 mortality included vaccination rates, new case rates, containment stringency, and healthcare capacity. When analyzed separately across Asia, Europe, and North America, Random Forest consistently outperformed the other models. Conclusion: Ensemble ML methods, particularly Random Forest, demonstrated superior predictive performance and adaptability across diverse settings. These findings highlight the potential of ML-driven epidemiological forecasting and suggest that ensemble approaches should be prioritized in pandemic preparedness and response efforts.