Interpretable Predictive Modeling for Smoking and Drinking Behavior using SHAP and LIME

Smoking and excessive alcohol consumption remain pressing public health issues. This research aimed to develop Machine Learning (ML) models leveraging sensor and biomarker data to predict smoking and drinking behaviors for targeted interventions. The Smoking and Drinking dataset (sourced from Kaggle) created from wearable sensor readings, blood tests, and lifestyle self-reports is used to train the 10 Machine Learning (ML) algorithms and validated for smoking and drinking prediction. The Random Forest model outperformed others, reaching an accuracy of 79.65% in predicting smoking, while the XGBoost model achieved a notable 73.96% accuracy in predicting drinking status. Overall, the ML models demonstrate strong capabilities for real-time smoking and drinking prediction using biological data. Such predictive analytics hold promise for early risk detection, personalized care, and nuanced interventions by healthcare systems. This research investigated ML algorithms to uncover and analyze patterns in lifestyle behaviors and associated health outcomes. Additionally, this work unlocks the Blackbox nature of ML models using Explainable AI tools “SHAP (SHapley Additive exPlanations)” and “LIME (Local interpretable model-agnostic explanations)”.