Scholarlink

<p>Accurately diagnosing early-stage heart disease poses a significant challenge in the medical field. Modern methods have proven beneficial for timely detection. This research addresses this challenge by applying four prominent machine learning algorithms: random forest classification, gradient boosting, AdaBoost, and light gradient boosting. The analysis incorporates the UCI Cleveland dataset (dataset I) and the comprehensive IEEE data port heart disease dataset (dataset II). The gradient boosting ML classifier, optimized through grid searchcv with the dynamic weighted ensemble and 10-fold cross-validation, attains a testing accuracy of 95 % On dataset I and 89.5 % On dataset II. The study introduces the novel use of grid search cross-validation to enhance training and testing processes. It implements a voting technique on both datasets to improve model accuracy predictions. Notably, when applying a Hard voting ensemble to dataset I and dataset II, the overall model attains an accuracy of 93.4 % and 91 %, respectively. While the gradient boosting classifier achieves the highest individual accuracy, the hard voting ensemble focuses on improving robustness and generalization by aggregating predictions from multiple classifiers. Furthermore, SHapley Additive exPlanations (SHAP) were applied to the gradient boosting classifier on dataset I to provide a clear and interpretable understanding of how individual features influence the model’s predictions. Features with widespread SHAP values, such as ’ca (chest pain type)’ and ’cp’ significantly affect predictions. Conversely, features like ’fbs’(fasting blood sugar) and ’restecg’(resting electrocardiographic results) show clusters near the center, suggesting a minimal impact on outcomes. This analysis provides a concise view of which features are most influential and their directional impact, enhancing our understanding of the model’s decision-making process.</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

Enhancing performance evaluation and ensemble methodology for early detection of heart disease with SHAP analysis

  • 期刊论文

  • 发表日期：

    2026年2月28日

查看全文
  • Nadikatla Chandrasekhar,
  • Samineni Peddakrishna,
  • Sreedhar Kollem

摘要

Accurately diagnosing early-stage heart disease poses a significant challenge in the medical field. Modern methods have proven beneficial for timely detection. This research addresses this challenge by applying four prominent machine learning algorithms: random forest classification, gradient boosting, AdaBoost, and light gradient boosting. The analysis incorporates the UCI Cleveland dataset (dataset I) and the comprehensive IEEE data port heart disease dataset (dataset II). The gradient boosting ML classifier, optimized through grid searchcv with the dynamic weighted ensemble and 10-fold cross-validation, attains a testing accuracy of 95 % On dataset I and 89.5 % On dataset II. The study introduces the novel use of grid search cross-validation to enhance training and testing processes. It implements a voting technique on both datasets to improve model accuracy predictions. Notably, when applying a Hard voting ensemble to dataset I and dataset II, the overall model attains an accuracy of 93.4 % and 91 %, respectively. While the gradient boosting classifier achieves the highest individual accuracy, the hard voting ensemble focuses on improving robustness and generalization by aggregating predictions from multiple classifiers. Furthermore, SHapley Additive exPlanations (SHAP) were applied to the gradient boosting classifier on dataset I to provide a clear and interpretable understanding of how individual features influence the model’s predictions. Features with widespread SHAP values, such as ’ca (chest pain type)’ and ’cp’ significantly affect predictions. Conversely, features like ’fbs’(fasting blood sugar) and ’restecg’(resting electrocardiographic results) show clusters near the center, suggesting a minimal impact on outcomes. This analysis provides a concise view of which features are most influential and their directional impact, enhancing our understanding of the model’s decision-making process.