Comparative Evaluation of Machine Learning Algorithms for Forecasting Infectious Diseases: Insights from COVID-19 and Dengue Data

This study evaluates the effectiveness of various machine learning (ML) models in forecasting COVID-19 case counts and predicting dengue outbreaks. Using publicly available datasets containing epidemiological data, climate variables, human mobility trends, and policy indicators, we trained and tested five ML algorithms: XGBoost, Random Forest, LSTM, SVM, and Logistic Regression. Our results demonstrate that XGBoost outperformed all other models, achieving the lowest mean absolute error (MAE = 1,079.2), root mean squared error (RMSE = 1,361.9), and the highest R² score (0.876) for COVID-19 forecasting. For dengue classification, XGBoost also led with the highest accuracy (91.3%), precision (88.7%), recall (90.8%), F1-score (89.7%), and ROC-AUC (0.949). Feature importance analysis confirmed that previous case counts, rainfall, humidity, vaccination rates, and mobility indices were the most influential variables. In terms of real-time application, XGBoost proved to be the most scalable and interpretable model, combining predictive strength with practical usability. These findings suggest that machine learning—particularly ensemble methods like XGBoost—can provide accurate, reliable, and real-time tools for infectious disease surveillance and early warning systems.