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

<b>Background.</b> 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. <b>Methods.</b> 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. <b>Results.</b> 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. <b>Conclusions.</b> 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.