Fusion Network for Enhanced Dengue Prediction Using Ensemble Methods and Dual Long Short-Term Memory Networks

Dengue fever poses a significant health threat that is expected to worsen due to climate change. Dengue epidemic prediction research faces challenges such as nonlinear interactions, model transparency issues, and biases in input data like temperature and reported cases. Existing dataset limitations hinder the full capture of transmission dynamics, reducing model accuracy. Overfitting, exacerbated by poor regularization or large ensemble sizes, further impacts prediction reliability. To address these challenges, a novel Advanced Fusion Ensemble Network with Dual Long Short-Term Memory that leverages advanced artificial intelligence techniques for enhanced predictive accuracy. Our approach integrates data preprocessing using the Adaptive Impute Scaler to ensure data consistency and reduce bias. Feature selection is achieved through a Gradual Ascent Network (GAN) to mitigate overfitting by extracting hierarchical features and reducing dimensionality. The Dual Long Short-Term Memory (D-LSTM) architecture effectively captures dengue transmission patterns. The model’s predictions are refined using a fully connected softmax function. The Local Epidemics Dengue Fever dataset was utilized, achieving a Root Mean Square Error (RMSE) of 52.44 and a Mean Absolute Error (MAE) of 25.45 for San Juan, and MAE and RMSE of 1.23 and 11.84, respectively, for Iquitos. Our findings demonstrate that the proposed AI-driven model offers accurate dengue forecasting and could be applicable to similar infectious diseases.