A Heterogeneous Ensemble Forecasting Model for Disease Prediction.

Area of Science:

• Epidemiology
• Machine Learning
• Time Series Analysis

Background:

• Accurate disease incidence forecasting is crucial for public health resource allocation.
• Existing ensemble models face challenges with accuracy, overfitting, and overdrift.

Purpose of the Study:

• To introduce a novel bragging-based ensemble forecasting model.
• To enhance prediction accuracy and reduce overfitting and overdrift in disease incidence data.
• To validate the model's performance on real-world disease datasets.

Main Methods:

• Data preprocessing using log and z-score transformation.
• Development and application of a bragging-based ensemble forecasting model.
• Comparative analysis against dynamic ensemble for time series, arbitrated dynamic ensemble, and random forest models.

Main Results:

• The proposed model demonstrated significant reductions in Mean Absolute Error (MAE) for tuberculosis (27.18%), dengue (3.07%), food poisoning (11.58%), and chickenpox (13.46%).
• The model consistently outperformed existing ensemble methods across all tested disease datasets.
• Achieved enhanced accuracy and robustness in disease incidence prediction.

Conclusions:

• The bragging-based ensemble model offers a superior approach for disease incidence forecasting.
• The model's effectiveness is validated on diverse disease datasets, showing improved predictive performance.
• This method provides a promising tool for epidemiological surveillance and public health planning.