Scholarlink

Hypertension is a major global health concern and a leading contributor to cardiovascular diseases. Accurate early detection is essential for timely intervention and management. Previous studies using the National Health and Nutrition Examination Survey (NHANES) data have largely focused on demographic and questionnaire-based features, potentially limiting predictive performance. This study proposes an extended approach to hypertension prediction by incorporating features from three NHANES data categories: demographic, examination, and laboratory data. Data from the 2013–2014, 2015–2016, and 2017–2018 cycles were combined to form a comprehensive dataset. 197 features were considered, and Analysis of Variance (ANOVA)-based feature selection was applied to identify the 15 most significant predictors, primarily related to body fat distribution and anthropometric measurements. The classification of hypertensive and normal individuals was based on the updated threshold of 130/80 mm Hg as per recent clinical guidelines. Multiple machine learning models including Logistic Regression, Decision Tree, SVM, Random Forest, and K-Nearest Neighbors (KNN) were trained using both train-test split and Stratified K-Fold cross-validation to ensure fair evaluation on the imbalanced dataset. Logistic Regression demonstrated the most consistent and superior performance in terms of accuracy with 84.02% and Area Under the Curve (AUC) Score of 90.1%. These findings highlight the importance of incorporating diverse clinical features particularly fat distribution metrics such as android and gynoid fat regions and Inflation level for improving hypertension prediction. Such physiological indicators, often overlooked in previous studies, show strong association with hypertension risk. Our results demonstrate that leveraging examination and laboratory data alongside traditional demographics can significantly enhance predictive performance in hypertension risk assessment. External validation with the 2011–2012 NHANES cycle achieved an AUC of 91.01%, confirming the relevance of selected features.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Hypertension Detection Using ANOVA-Correlation Feature Selection with Updated Diagnostic Thresholds Using Machine Learning Approach

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Mirajkumar Malam,
  • Jigna Jadav

摘要

Hypertension is a major global health concern and a leading contributor to cardiovascular diseases. Accurate early detection is essential for timely intervention and management. Previous studies using the National Health and Nutrition Examination Survey (NHANES) data have largely focused on demographic and questionnaire-based features, potentially limiting predictive performance. This study proposes an extended approach to hypertension prediction by incorporating features from three NHANES data categories: demographic, examination, and laboratory data. Data from the 2013–2014, 2015–2016, and 2017–2018 cycles were combined to form a comprehensive dataset. 197 features were considered, and Analysis of Variance (ANOVA)-based feature selection was applied to identify the 15 most significant predictors, primarily related to body fat distribution and anthropometric measurements. The classification of hypertensive and normal individuals was based on the updated threshold of 130/80 mm Hg as per recent clinical guidelines. Multiple machine learning models including Logistic Regression, Decision Tree, SVM, Random Forest, and K-Nearest Neighbors (KNN) were trained using both train-test split and Stratified K-Fold cross-validation to ensure fair evaluation on the imbalanced dataset. Logistic Regression demonstrated the most consistent and superior performance in terms of accuracy with 84.02% and Area Under the Curve (AUC) Score of 90.1%. These findings highlight the importance of incorporating diverse clinical features particularly fat distribution metrics such as android and gynoid fat regions and Inflation level for improving hypertension prediction. Such physiological indicators, often overlooked in previous studies, show strong association with hypertension risk. Our results demonstrate that leveraging examination and laboratory data alongside traditional demographics can significantly enhance predictive performance in hypertension risk assessment. External validation with the 2011–2012 NHANES cycle achieved an AUC of 91.01%, confirming the relevance of selected features.