Enhancing Influenza-Like Illness forecasting: An ensemble approach combining mathematical and deep learning models amidst the COVID-19 pandemic

Background: Timely and accurate short-term forecasting of Influenza-Like Illness (ILI) is crucial for guiding outbreak response, optimizing healthcare resource allocation, and informing public health interventions. The COVID-19 pandemic, which disrupted seasonal ILI dynamics due to widespread nonpharmaceutical interventions (NPI), underscored the urgent need for adaptive and reliable forecasting frameworks. Method: In this study, we present a novel ensemble modeling approach that combines a mechanistic n-subepidemic model with a Monte Carlo Dropout Long Short-Term Memory (LSTM) neural network to improve age-specific ILI forecasting performance in South Korea. By capturing both the structured dynamics of disease spread and nonlinear temporal dependencies, our ensemble method adapts to pandemic-altered transmission patterns while offering robust uncertainty quantification. Age-stratified forecasting allows the framework to capture heterogeneity in vulnerability and transmission across demographic groups, providing more targeted insights for policy and planning. Result: We evaluated forecasting performance across four epidemic waves using the Weighted Interval Score (WIS), Mean Absolute Error (MAE), consistently finding that the ensemble models outperformed individual approaches. Conclusion: These findings underscore the power of hybrid forecasting approaches to improve epidemic preparedness and response, providing a flexible data-driven framework that can evolve with changing transmission dynamics and extend to other emerging infectious threats.