Multi-Organ Impact of Smoking Using Explainable Artificial Intelligence: Prediction, Validation, and Organ Impact Pattern

This study aims to investigate the multi-organ impact of smoking using an Explainable Artificial Intelligence (XAI)-based framework by analyzing liver, kidney, and metabolic health indicators from the Kaggle Body Signal of Smoking dataset containing 55,692 patient records and 27 clinical features. The objective is to identify smoking-associated biomarkers and infer a potential organ impact pattern using machine learning and explainability techniques. Data preprocessing included removal of missing values, label encoding of categorical variables, and separation of features (X) and target variable (y = smoking). The dataset was split into 80% training and 20% testing sets. Three classification models - Random Forest, XGBoost, and Logistic Regression-were trained, with hyperparameter tuning performed using GridSearchCV on Random Forest. Model validation included Accuracy, F1-score, Receiver Operating Characteristic Area Under the Curve (ROC-AUC), 5-fold cross-validation, independent sample t-tests, and SHapley Additive exPlanations (SHAP) for interpretability. Results showed that Random Forest achieved the best predictive performance with strong classification accuracy and ROC-AUC. Statistical testing revealed significant differences (P < 0.05) in Gamma-Glutamyl Transferase (GTP), Alanine Aminotransferase (ALT), serum creatinine, hemoglobin, and triglycerides between smokers and non-smokers. SHAP and feature importance analysis consistently identified GTP, ALT, and serum creatinine as major predictors. This study suggests a potential organ impact pattern as Liver > Kidney > Metabolic System based on statistical analysis and XAI techniques.