Scholarlink

<p>Early diagnosis of cardiovascular diseases (CVD) is crucial for improving patient survival and reducing healthcare costs. However, the high cost, time-consuming nature of traditional diagnostic methods, and the need for expert interpretation make it difficult to expand early diagnosis, especially in regions with limited healthcare services. In this study, a novel hybrid optimization approach based on Grey Wolf Optimization (GWO), Whale Optimization Algorithm (WOA), and Arithmetic Optimization Algorithm (AOA) is proposed to optimize the hyperparameters of a deep convolutional neural network (CNN) model. The proposed hybrid method operates sequentially in three stages: exploration with GWO, exploitation with WOA, and fine-tuning with AOA. In this study, heart disease prediction is performed using the deep CNN model optimized with the proposed hybrid optimization method on a comprehensive and heterogeneous heart disease dataset (1,190 patient records) generated by merging the Statlog, Long Beach VA, Switzerland, Hungary, and Cleveland datasets. Nine critical hyperparameters, including the number of filters, kernel size, pooling size, fully connected layer neurons, dropout rate, learning rate, mini-batch size, optimizer type, and maximum epoch number, are systematically optimized. The proposed method overcomes the challenges encountered in data integration, preprocessing, and hyperparameter optimization, and provides a prediction model with high accuracy and clinical applicability. Experimental results show that the proposed hybrid optimized deep CNN approach achieves higher accuracy (96.22%), F1 score (96.41%), sensitivity (96.03%), and precision (96.80%) values compared to existing studies. Statistical validation using the Wilcoxon signed-rank test confirmed the significance of the performance improvements (<i>p</i> = 0.016). SHAP analysis revealed that ST slope, chest pain type, and oldpeak parameters were the most effective features for prediction. Data augmentation performed with SMOTE improved the performance to 97.42% accuracy on the double-balanced dataset. Analyses show that the proposed method provides a practical, explainable and statistically validated tool for early prediction of cardiovascular diseases and helps healthcare professionals make informed decisions.</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

A novel hybrid metaheuristic for optimizing deep CNN hyperparameters to enhance heart disease prediction on a comprehensive merged dataset

  • 期刊论文

  • 发表日期：

    2026年1月16日

查看全文
  • Timur Lale,
  • Gökhan Yüksek,
  • Rıdvan Fırat Çınar

摘要

Early diagnosis of cardiovascular diseases (CVD) is crucial for improving patient survival and reducing healthcare costs. However, the high cost, time-consuming nature of traditional diagnostic methods, and the need for expert interpretation make it difficult to expand early diagnosis, especially in regions with limited healthcare services. In this study, a novel hybrid optimization approach based on Grey Wolf Optimization (GWO), Whale Optimization Algorithm (WOA), and Arithmetic Optimization Algorithm (AOA) is proposed to optimize the hyperparameters of a deep convolutional neural network (CNN) model. The proposed hybrid method operates sequentially in three stages: exploration with GWO, exploitation with WOA, and fine-tuning with AOA. In this study, heart disease prediction is performed using the deep CNN model optimized with the proposed hybrid optimization method on a comprehensive and heterogeneous heart disease dataset (1,190 patient records) generated by merging the Statlog, Long Beach VA, Switzerland, Hungary, and Cleveland datasets. Nine critical hyperparameters, including the number of filters, kernel size, pooling size, fully connected layer neurons, dropout rate, learning rate, mini-batch size, optimizer type, and maximum epoch number, are systematically optimized. The proposed method overcomes the challenges encountered in data integration, preprocessing, and hyperparameter optimization, and provides a prediction model with high accuracy and clinical applicability. Experimental results show that the proposed hybrid optimized deep CNN approach achieves higher accuracy (96.22%), F1 score (96.41%), sensitivity (96.03%), and precision (96.80%) values compared to existing studies. Statistical validation using the Wilcoxon signed-rank test confirmed the significance of the performance improvements (p = 0.016). SHAP analysis revealed that ST slope, chest pain type, and oldpeak parameters were the most effective features for prediction. Data augmentation performed with SMOTE improved the performance to 97.42% accuracy on the double-balanced dataset. Analyses show that the proposed method provides a practical, explainable and statistically validated tool for early prediction of cardiovascular diseases and helps healthcare professionals make informed decisions.