Explainable Machine Learning Framework for Predicting Cardiometabolic Risk Using Meal Timing and Eating Habits

Abstract Background Eating timing and regularity represent new contributors to metabolic health, however, the time-based aspect of eating behavior is rarely incorporated into traditional cardiometabolic risk assessment strategies. Objectives The aim of this study was to create an explainable machine learning (ML) model to predict cardiometabolic risk based on anthropometric, dietary and chrono-nutritional data. Methodology This cross-sectional study included 300 adults for whom demographic, waist-hip ratio (WHR), Body Mass Index (BMI), blood pressure, eating timing (earliest time, latest time, meal frequency) and eating behavior observations were recorded. Nested 10-fold cross-validation and hyperparameter tuning were used to create three models (Logistic Regression (LR), Random Forest (RF), XGBoost) with interpretability established through Shapley Additive Explanations (SHAP). Discrimination was evaluated with AUC-ROC, accuracy, precision, recall, F1-score and Brier score. Results XGBoost was the best model compared to RF (AUC = 0.94) and LR (AUC = 0.92) (AUC-ROC = 0.98, accuracy = 93%). Later dinner timing (OR = 2.5), irregular meal timing (OR = 1.6), and reduced adherence to the recommended meal frequency (OR = 1.8) were the top independent predictors of cardiometabolic danger; conversely In contrast, an increased frequency of meals and an awareness of the significance of meal timing positively impacted cardiometabolic danger. SHAP further interpreted XGBoost to conclude that BMI and WHR were the most significant predictors in addition to meal regularity. Conclusion Explainable ML modeling from mealtimes and nutritional behaviors provide accurate and interpretable prediction of threat. Improved awareness of meal timing serves as a modifiable behavioral intervention in preventative cardiometabolic efforts.