Enhanced Smoking Behavior Prediction Using Stacked Ensemble Learning and Biometric Indicators

Despite widespread public health campaigns, smoking remains a major contributor to chronic illnesses such as heart disease, stroke, respiratory conditions, and various cancers. This persistent issue arises from the complex interplay of factors influencing smoking behavior, including nicotine addiction, mental health conditions, and social influences. Accurately understanding and predicting smoking habits is crucial for reducing the burden of tobacco-related diseases. Targeted identification of individuals at higher risk enables healthcare providers and policymakers to design more effective interventions and allocate resources efficiently, ultimately improving population health outcomes. In this study, we propose a hybrid predictive model that combines machine learning and ensemble stacking techniques to enhance the accuracy of smoking status classification. The model leverages various biometric and clinical variables such as age, height, weight, waist circumference, blood pressure, blood glucose, cholesterol, triglycerides, and liver enzymes. We evaluated several algorithms, including Extra Trees, Random Forest, XGBoost, SVC, Bagging, LGBM, CatBoost, and neural networks, integrating the top three into a meta-model. This approach achieved an accuracy of ۸۴%, highlighting its potential to support clinical decision-making and public health planning.

Keywords:

Smoking , Behavior Prediction , Machine Learning , Ensemble Models , Biosignals , Public Health