Scholarlink

<p>Heart disease remains the leading cause of death worldwide, underscoring the urgent need for accurate, timely diagnosis to improve patient outcomes. Traditional diagnostic methods often struggle with accuracy, delayed intervention, and the complexity of analyzing diverse patient data. Moreover, existing machine learning approaches face challenges due to missing values, class imbalances, and limited features, all of which can negatively affect model performance. To overcome these limitations, this study introduces a novel autoencoder-based framework to enhance feature engineering for heart disease prediction. Autoencoders are employed for both dimensionality reduction and the generation of new, informative features that capture complex, non-linear relationships within the data. The methodology is evaluated using eight diverse heart disease datasets, including the widely recognized Cleveland and Hungarian datasets, with performance assessed through accuracy, balanced accuracy, ROC AUC, F1 score, and computational efficiency. Experimental results demonstrate notable improvements, with the method achieving 91.12% accuracy on the Cleveland dataset and 86.96% on the Hungarian dataset. Comparative analysis across four experimental scenarios—baseline performance with original features, autoencoder-based feature generation, feature reduction, and Bayesian hyperparameter optimization—reveals that the proposed approach substantially outperforms traditional techniques, achieving up to 96% accuracy on concatenated datasets and 95% on a comprehensive dataset. These compelling results highlight the potential of advanced autoencoder-driven feature engineering combined with sophisticated optimization strategies to significantly enhance diagnostic accuracy and generalization in heart disease prediction.</p>

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

An autoencoder-based approach for feature engineering in cardiovascular disease prediction

  • 期刊论文

  • 发表日期：

    2025年8月22日

查看全文
  • Amr E. Eldin Rashed,
  • Mahmoud Badawy,
  • Mostafa A. Elhosseini,
  • Waleed M. Bahgat

摘要

Heart disease remains the leading cause of death worldwide, underscoring the urgent need for accurate, timely diagnosis to improve patient outcomes. Traditional diagnostic methods often struggle with accuracy, delayed intervention, and the complexity of analyzing diverse patient data. Moreover, existing machine learning approaches face challenges due to missing values, class imbalances, and limited features, all of which can negatively affect model performance. To overcome these limitations, this study introduces a novel autoencoder-based framework to enhance feature engineering for heart disease prediction. Autoencoders are employed for both dimensionality reduction and the generation of new, informative features that capture complex, non-linear relationships within the data. The methodology is evaluated using eight diverse heart disease datasets, including the widely recognized Cleveland and Hungarian datasets, with performance assessed through accuracy, balanced accuracy, ROC AUC, F1 score, and computational efficiency. Experimental results demonstrate notable improvements, with the method achieving 91.12% accuracy on the Cleveland dataset and 86.96% on the Hungarian dataset. Comparative analysis across four experimental scenarios—baseline performance with original features, autoencoder-based feature generation, feature reduction, and Bayesian hyperparameter optimization—reveals that the proposed approach substantially outperforms traditional techniques, achieving up to 96% accuracy on concatenated datasets and 95% on a comprehensive dataset. These compelling results highlight the potential of advanced autoencoder-driven feature engineering combined with sophisticated optimization strategies to significantly enhance diagnostic accuracy and generalization in heart disease prediction.