AI-Powered Opioid Overdose Risk Prediction and Prevention Using Explainable Machine Learning with Policy Recommendations

Public health problems such as the growing opioid crisis require feasible prediction tools for prevention and intervention. This study proposes an AI-based framework for measuring the risk of opioid overdose using machine learning models and explainable AI approaches. Due to the lack of available opioid-specific datasets, this study uses the PIMA Indian Diabetes dataset in a repurposed simulation framework to demonstrate methodological feasibility. The proposed approach reused a Random Forest Classifier to predict simulated opioid overdose risk and achieved an accuracy of 82 %, an F1-score of 0.79, and an ROC-AUC of 0.81, reflecting strong model performance. SHAP-based explainability was incorporated to highlight the relative importance of predictors such as Glucose, Age, BMI, and Insulin, enabling users to interpret model outputs. It defined and described key evaluation metrics including precision, recall, F1-score, and ROC AUC applied to the models. Overall, the study supports the integration of explainable ML into overdose risk modeling and offers model-informed insights for public health intervention planning. Future research will focus on applying this framework to real opioid-related datasets for improved clinical decision-making.