Scholarlink

After the COVID-19 pandemic, Cardiovascular Disease (CVD) has arisen as a notable health concern in middle-aged individuals. Discovering the CVD earlier and taking necessary precautions is much needed to reduce the risk of CVD complications [1]. This study explores various reinforcement learning (RL) techniques to improve CVD forecasting among people aged 40 to 60 years. We present a novel framework in its integration of customizable actions within reinforcement learning (RL), allowing dynamic ensemble selection that adapts to real-time feedback. This approach differs from traditional static methods by continuously optimizing classifier selection based on evolving data. We used preprocessing techniques such as label encoding, one-hot encoding, and numerical feature standardization by employing a heart disease dataset [2] comprising 696 rows and 14 attributes. Hyperparameter tuning was conducted using GridSearchCV across 10 classifiers. Classifier performance was evaluated through cross-validation with metrics including accuracy, precision, recall, F1 score, ROC AUC score, and average precision score. The RL framework dynamically selected ensemble strategies by majority voting, weighted voting, and best-performing classifier from different classifiers to balance exploration and exploitation using techniques like Upper Confidence Bound (UCB), epsilon-greedy, Thompson Sampling, and SoftMax. The comparisons and observations prove that Thompson Sampling performed the best with an optimized accuracy of 98% and 100% precision. The results highlight the effectiveness of RL-based dynamic ensemble selection in enhancing prediction for middle-aged people at risk of CVD, concluding a reliable forecasting system for improving early detecting techniques and preventing CVD prior to its occurrences [3].

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 Cardiovascular Disease Prediction Among Middle-Aged Individuals Using Reinforcement Learning Dynamic Ensemble Selection with Customizable Actions and Exploration-Exploitation Balance

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • S. R. Divyasri,
  • A. Rama Prasath

摘要

After the COVID-19 pandemic, Cardiovascular Disease (CVD) has arisen as a notable health concern in middle-aged individuals. Discovering the CVD earlier and taking necessary precautions is much needed to reduce the risk of CVD complications [1]. This study explores various reinforcement learning (RL) techniques to improve CVD forecasting among people aged 40 to 60 years. We present a novel framework in its integration of customizable actions within reinforcement learning (RL), allowing dynamic ensemble selection that adapts to real-time feedback. This approach differs from traditional static methods by continuously optimizing classifier selection based on evolving data. We used preprocessing techniques such as label encoding, one-hot encoding, and numerical feature standardization by employing a heart disease dataset [2] comprising 696 rows and 14 attributes. Hyperparameter tuning was conducted using GridSearchCV across 10 classifiers. Classifier performance was evaluated through cross-validation with metrics including accuracy, precision, recall, F1 score, ROC AUC score, and average precision score. The RL framework dynamically selected ensemble strategies by majority voting, weighted voting, and best-performing classifier from different classifiers to balance exploration and exploitation using techniques like Upper Confidence Bound (UCB), epsilon-greedy, Thompson Sampling, and SoftMax. The comparisons and observations prove that Thompson Sampling performed the best with an optimized accuracy of 98% and 100% precision. The results highlight the effectiveness of RL-based dynamic ensemble selection in enhancing prediction for middle-aged people at risk of CVD, concluding a reliable forecasting system for improving early detecting techniques and preventing CVD prior to its occurrences [3].