Enhancing Hypertension Risk Diagnosis Using a Hybrid MachineLearning Framework: Leveraging Body Composition Data

Abstract

Hypertension, widely recognized as the “silent killer,” remains a leading cause of cardiovascular, renal, and neurological complications worldwide. This study proposes a dual-scenario hybrid machine learning framework for hypertension risk prediction using noninvasive body composition features, aimed at enhancing both interpretability and predictive reliability. In Scenario 1, an unsupervised clustering analysis inspired by self-labeling principles was performed exclusively on hypertensive individuals, where five physiological subgroups were identified via K-Means clustering and validated using Silhouette (0.3371),Davies–Bouldin (1.0094), and Calinski–Harabasz (720.10) indices. Significant inter cluster variability (p < 0 001) was observed across key indicators such as FATP, RLFATP, LLFATP, FATM, and age. Among the tested models, the support vector machine (SVM) with random oversampling achieved the best performance (accuracy = 99 08%, F1 = 98 04%, AUC = 99 98%),confirming effective subgroup discrimination. In Scenario 2, a comprehensive binary classification between healthy and hypertensive subjects was conducted using five models—Extra Trees, KNN, SVM, Gaussian Naive Bayes, and Decision Tree—across multiple configurations. The cluster-augmented dataset yielded the best results, with the Extra Trees classifier achieving superior performance (accuracy = 98 23%, recall = 98 30%, precision = 98 17%, F1 = 98 23%, AUC = 99 87%).Clustering and feature selection both improved generalization, particularly for ensemble-based learners. Overall, Scenario 2demonstrated the highest predictive accuracy and stability, whereas Scenario 1 provided valuable interpretability through subgroup discovery. These findings highlight that integrating unsupervised clustering with supervised classification offers a robust and explainable framework for personalized hypertension risk prediction, contributing to early detection and precisionhealthcare.