An Explainable Machine Learning-Based Integrated Predictive Framework for Early Risk Stratification of Lifestyle Diseases Using Multi-Source Healthcare Data

Lifestyle diseases -diabetes mellitus, cardiovascular disease, and hypertension -represent a rapidly growing global health burden attributable to sedentary behavior, chronic stress, and poor dietary practices.Early risk identification enables timely preventive intervention and reduces long-term healthcare expenditure.This study proposes an explainable machine learning-based integrated predictive framework for early risk stratification of lifestyle diseases using a multisource health dataset, where "multi-source" denotes the integration of distinct clinical (e.g., blood pressure, cholesterol, body mass index) and behavioral-lifestyle (e.g., smoking status, physical activity, diet) indicator domains jointly captured within the CDC Behavioral Risk Factor Surveillance System survey instrument, rather than merging multiple independent databases.The balanced binary variant of the Diabetes Health Indicators Dataset (N = 70,692; 50:50 class distribution) was used for model development.Logistic regression served as the interpretable baseline classifier, and random forest served as the primary ensemble model.Model performance was assessed via five-fold stratified cross-validation and held-out test evaluation using accuracy, precision, recall, F1-score, area under the receiver operating characteristic curve (AUC), and average precision.McNemar's test and paired t-tests were applied to confirm the statistical significance of performance differences.Logistic regression achieved a cross-validated AUC of 0.820 (±0.006) and a held-out test AUC of 0.818, yielding superior precision (p < 0.05).Random forest achieved a cross-validated recall of 0.788 (±0.008) and a heldout recall of 0.784, yielding superior sensitivity for positive-case detection (p < 0.05).SHapley Additive exPlanations analysis identified general health status, body mass index, high blood pressure, and age as the strongest predictors.Risk stratification partitioned the test population into three clinically interpretable tiers.The integrated explainable framework demonstrates viability for actionable, transparent healthcare decision support in preventive medicine.