Nonlinear Enhanced Prediction of Influenza-Like Illness Using Seasonal-Trend Decomposition and Temporal Dependency

Influenza-like illness (ILI) is a common respiratory infectious disease that puts significant strain on medical resources and leads to numerous fatalities worldwide. Accurate prediction of ILI incidence is crucial for public health. However, existing predictive models struggle to capture non-linear patterns due to data limitations and noise, leading to reduced accuracy. To tackle this challenge, we introduce the Seasonal-trend Decomposition and Temporal Dependency Nonlinear Enhanced (STDNE) model. First, STDNE employs a seasonal-trend decomposition enhanced model to separate trend and seasonal components, reducing residual noise and addressing data sparsity. Second, it integrates a temporal dependency modeling technique to capture long-term relationships among variables. Finally, a nonlinear pattern enhancement component is introduced to model complex interactions within the data. Experimental results on three small-sample ILI case datasets with weekly statistics show that STDNE reduces MAE, MSE, and RMSE by 13.62%, 23.61%, and 14.00% on average compared to traditional models, and by 18.85%, 11.49%, and 9.50% compared to Transformer-based models, demonstrating its superior accuracy in ILI case prediction.