Machine Learning for Environmental Antimicrobial Resistance Risk Prediction

Antimicrobial resistance (AMR) poses a major global health threat, and environmental reservoirs play a critical role in the dissemination of antimicrobial resistance genes (ARGs). This study proposes a machine learning (ML) framework to predict environmental ARG dissemination risk using real-world surveillance data. Gene-level, environmental, spatial, and abundance-related features were integrated into a supervised learning pipeline. Nine ML models were systematically evaluated, including Logistic Regression, Random Forest, Gradient Boosting, XGBoost, LightGBM, and neural networks. Ensemble-based models demonstrated superior performance, with LightGBM achieving the highest accuracy (98.96%) and F1-score (97.24%). Feature importance and SHAP-based explainability analyses identified ARG abundance, habitat type, and geographic attributes as dominant drivers of dissemination risk. The findings demonstrate that ML can effectively capture complex nonlinear ARG-environment interactions and provide interpretable, risk-oriented predictions. The proposed framework offers a scalable approach for environmental AMR surveillance and supports integration into digital health systems within a One Health context.