Scholarlink

Timely diagnosis and prevention of cardiovascular disease (CVD) depend on accurate risk prediction. This study integrates statistical hypothesis testing with machine learning classifiers to enhance predictive performance and interpretability. Using the Framingham Heart Study (4,240 samples) and Heart Failure Clinical Records (299 samples), independent t-tests, chi-square tests, and correlation analysis were used to select significant features. Feature selection improved Support Vector Machine (SVM) test accuracy from 0.58 to 0.72 for the Heart Failure dataset and maintained accuracy at 0.84 for the Framingham dataset. Hypothesis testing revealed gender-specific effects of BMI and education on CHD risk, confirming or disproving widely held clinical beliefs. To guarantee accessibility for real-world use, an interactive user interface was created. The findings show that hypothesis-driven feature selection can find clinically significant insights, maintain predictive performance, and simplify models.

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

Cardiovascular Disease Risk Prediction Using Hypothesis-Based Feature Selection and Machine Learning

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Lohitha Kanisettypalli,
  • Harsda Shrivastava,
  • Jiya Borikar,
  • Neetu Srivastava

摘要

Timely diagnosis and prevention of cardiovascular disease (CVD) depend on accurate risk prediction. This study integrates statistical hypothesis testing with machine learning classifiers to enhance predictive performance and interpretability. Using the Framingham Heart Study (4,240 samples) and Heart Failure Clinical Records (299 samples), independent t-tests, chi-square tests, and correlation analysis were used to select significant features. Feature selection improved Support Vector Machine (SVM) test accuracy from 0.58 to 0.72 for the Heart Failure dataset and maintained accuracy at 0.84 for the Framingham dataset. Hypothesis testing revealed gender-specific effects of BMI and education on CHD risk, confirming or disproving widely held clinical beliefs. To guarantee accessibility for real-world use, an interactive user interface was created. The findings show that hypothesis-driven feature selection can find clinically significant insights, maintain predictive performance, and simplify models.