Metabolic syndrome (MetS) is a complex condition characterized by risk factors such as central obesity, insulin resistance, hypertension, and dyslipidemia, significantly elevating the risk of cardiovascular diseases and type 2 diabetes. With a global prevalence of 20–25%, MetS presents a major public health challenge, compounded by the limitations of binary diagnostic criteria that fail to capture the continuum of metabolic dysfunction. This thesis proposes a novel machine learning framework for multi-level risk stratification of MetS, aiming to improve early detection and personalized interventions. By integrating advanced computational techniques, the framework provides a granular and interpretable assessment of metabolic health, addressing shortcomings in traditional diagnostic approaches. The methodology utilizes a publicly available dataset from the National Health and Nutrition Examination Survey (NHANES) with 2,401 individuals and 15 features, including demographic, clinical, and laboratory data. A three-tier risk classification (Low, Moderate, High) was established using gender-specific thresholds for key MetS markers like waist circumference and HDL cholesterol. Fuzzy C-means clustering with Gower distance enriched the dataset by generating membership and distance features to capture complex metabolic patterns. Multiple classifiers were assessed via 10-fold cross-validation, with the HistGradientBoostingClassifier outperforming others. Local Interpretable Model-agnostic Explanations (LIME) ensured clinical transparency by identifying key predictors driving risk classifications. The framework achieved exceptional performance, with the HistGradientBoostingClassifier yielding an accuracy of 99.36%, precision of 99.37%, recall of 99.36%, and AUC of 99.84%, surpassing state-of-the-art studies on the same dataset. Waist circumference, HDL cholesterol, blood glucose, and uric acid were identified as top predictors, aligning with clinical MetS criteria, while cluster-derived features enhanced predictive power by revealing latent risk profiles. LIME explanations offered actionable insights, enabling clinicians to understand patient-specific risk factors and tailor interventions. The framework’s multi-level classification and interpretability mark significant advancements, supporting precision medicine in MetS management. This research advances computational diagnostics by integrating fuzzy clustering, feature engineering, and explainability, setting a new benchmark for MetS risk assessment. Limitations include the static nature of the analysis, which overlooks temporal disease progression, and the exclusion of certain MetS components like blood pressure due to dataset constraints. Future work will explore longitudinal data to model risk trajectories and incorporate additional clinical features to enhance diagnostic accuracy. These advancements will further strengthen the framework’s potential to deliver scalable, interpretable solutions for managing complex metabolic disorders.
