Retrospective Analysis and Cross-Validated Forecasting of West Nile Virus Transmission in Italy: Insights from Climate and Surveillance Data.

Background. West Nile Virus (WNV) represents a significant public health concern in Europe, with Italy-particularly its northern regions-experiencing recurrent outbreaks. Climate variables and vector dynamics are known to significantly influence transmission patterns, highlighting the need for reliable predictive models to enable timely outbreak detection and response. Methods. We integrated epidemiological data on human WNV infections in Italy (2012-2024) with high-resolution climate variables (temperature, humidity, and precipitation). Using advanced feature engineering and a gradient boosting framework (XGBoost), we developed a predictive model optimized through time-series cross-validation. Results. The model achieved high predictive accuracy at the national level (R2 = 0.994, MAPE = 5.16%) and maintained robust performance across the five most affected provinces, with R2 values ranging from 0.896 to 0.996. SHAP analysis identified minimum temperature as the most influential climate predictor, while maximum temperature and rainfall demonstrated considerably weaker associations with case incidence. Conclusions. This machine learning approach provides a reliable framework for forecasting WNV outbreaks and supports evidence-based public health responses. The integration of climate and epidemiological data enhances surveillance capabilities and enables informed decision-making at regional and local levels.