Scopus Indexed Publications

Paper Details


Title
Ensemble Modeling for Global Forecasting of Monkeypox Incidence Using Statistical and Machine Learning Approaches

Author
Masuma Parvin, Sharmin Akther,

Email

Abstract

Accurate forecasting of Monkeypox outbreaks is essential for timely public health interventions and effective resource allocation globally. While numerous statistical and machine learning models have been applied to disease prediction, limited research has systematically compared ensemble approaches against individual models across multiple continents. This study addresses this gap by evaluating the forecasting performance of seven individual models Autoregressive Integrated Moving Average (ARIMA), exponential smoothing (ETS), Prophet, Artificial Neural Networks (ANNs), Support Vector Regression (SVR), Random Forest (RF), and Extreme Gradient Boosting (XGBoost)—alongside two ensemble strategies (Weighted and Equal Ensemble) across six continents using daily case data from November 2022 to April 2025. Model performance was assessed using Root Mean Square Error (RMSE), Mean Absolute Error (MAE), symmetric Mean Absolute Percentage Error (sMAPE), Mean Absolute Scaled Error (MASE), and first-order autocorrelation of residuals (ACF1). Results reveal that model performance is highly continent specific: ARIMA excelled in Africa (RMSE: 102.91) and North America (RMSE: 5.43), ETS performed best in Asia (RMSE: 11.05), Prophet dominated in South America (RMSE: 6.17), and XGBoost achieved the lowest error in Europe (RMSE: 42.43). The Equal Ensemble consistently outperformed the Weighted Ensemble across most continents, with substantial improvements of 30.36% in North America and 20.58% in South America. Future forecasts from July to December 2026 using both the best individual model and the Equal Ensemble indicate that monkeypox transmission is expected to remain low across most continents, with Africa showing the highest projected activity (774 cases/day). Computational complexity analysis confirmed the framework’s feasibility for real-time deployment, with all models training within 60 s and achieving inference latency under 0.1 s. These findings support continent-specific model selection and demonstrate that equal-weight ensemble averaging provides a robust, interpretable, and operationally feasible approach for global Monkeypox surveillance and public health preparedness. Future work should explore adaptive ensemble methods, stacking approaches, and integration of real-time data streams to further enhance forecasting capabilities for emerging infectious disease threats.


Keywords

Journal or Conference Name
Applied Computational Intelligence and Soft Computing

Publication Year
2026

Indexing
scopus