Integrating Statistical and Machine-Learning Approaches for Salmonella enterica Surveillance in Northwestern Italy: A One Health Data-Driven Framework

<i>Salmonella enterica</i> is a major cause of foodborne illness globally. We analysed 41,945 food samples collected under official surveillance in Piedmont (north-western Italy) between 2013 and 2023 to characterise contamination patterns and evaluate an integrated analytical framework combining classical statistical modelling with machine-learning prediction. Overall prevalence was low (2.20%; 95% CI: 2.06-2.35) but heterogeneous across matrices, with poultry and pork displaying the highest contamination levels (11.8% and 7.14%). Risk increased at distribution/retail stages, and contamination declined markedly from 2013 to 2018, with lower levels in late autumn. Meteorological factors had minimal influence. Mixed-effects models identified food category and production stage as the main determinants of contamination, while the XGBoost algorithm showed stable predictive performance (median absolute error ≈ 0.02) and spatially coherent estimates. SHAP analyses confirmed food composition variables as the dominant predictors. These findings highlight persistent vulnerabilities within poultry and swine supply chains, particularly at post-production stages, and illustrate the complementary value of combining explanatory and predictive approaches to strengthen risk-based, One Health-aligned food-safety surveillance.