Generalized additive model integrating multi-source data for short-term influenza forecasting in Shenzhen, China (2023–2025)

BackgroundThe COVID-19 pandemic has reshaped the global epidemiology of respiratory infectious diseases, posing new challenges for influenza forecasting. Existing studies are often limited by reliance on single data sources, poor interpretability, or failure to account for nonlinear relationships among variables, which restricts their ability to balance prediction accuracy and practical utility for public health decision-making. This study aimed to develop and validate a multisource data-integrated generalized additive model (GAM) to forecast influenza activity in Shenzhen, China.MethodsUsing surveillance and auxiliary data from 2023 to 2025, we developed GAM models incorporating local and Hong Kong influenza surveillance, cross-boundary mobility metric, meteorological factors, and Baidu Search Index data. The predictive performance of the GAM was compared with Seasonal Autoregressive Integrated Moving Average with Exogenous Variables (SARIMAX) model. Model accuracy was evaluated using root mean square error (RMSE), mean absolute percentage error (MAPE), and R2.ResultsThe multisource data-driven GAM exhibited high predictive accuracy across short-term forecasting horizons. For 1-week ahead forecasts, the model achieved an R2 of 0.85 (95% CI: 0.74–0.92). Notably, performance remained robust for 2- and 3-week forecasts, with R2 values of 0.80 (95% CI: 0.69–0.87) and 0.74 (95% CI: 0.62–0.83), respectively. The GAM demonstrated superior overall performance compared with SARIMAX.ConclusionThe multisource data-integrated GAM provides robust and stable influenza forecasts for Shenzhen up to 3 weeks in advance. This approach provides a valuable tool to support cross-boundary public health collaboration between Hong Kong and Shenzhen, and might serve as a reference for the development of broader regional public health strategies in future research.