Scholarlink

Cerebral stroke is a major global health concern, necessitating improved predictive models to enhance patient outcomes and healthcare efficiency. However, an ongoing challenge in stroke prediction is the significant class imbalance in datasets, where stroke instances are heavily underrepresented compared to non-stroke cases. This imbalance often results in biased machine learning models that perform poorly in predicting stroke occurrences. To address this issue, we evaluated three data-balancing methods: oversampling, undersampling, and the synthetic minority oversampling technique (SMOTE). Using a dataset comprising 42,617 healthy cases and 783 stroke cases, we assessed ten machine learning models, including logistic regression (LR), Naive Bayes (NB), support vector machine (SVM), decision tree (DT), random forest (RF), XGBoost, LightGBM, CatBoost, K-nearest neighbors (KNN), and artificial neural network (ANN). Our results indicate that combining SMOTE with the CatBoost model substantially enhances both prediction accuracy and sensitivity compared to using undersampling or oversampling alone. This underscores SMOTE’s effectiveness in addressing class imbalance in healthcare datasets. By highlighting the critical role of data balancing in improving prediction, this study offers practical insights for developing and clinically applicable predictive 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

Enhancing Stroke Prediction with Machine Learning Models and Data Balancing Techniques

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Hieu T. P. Le,
  • Luan N. T. Huynh

摘要

Cerebral stroke is a major global health concern, necessitating improved predictive models to enhance patient outcomes and healthcare efficiency. However, an ongoing challenge in stroke prediction is the significant class imbalance in datasets, where stroke instances are heavily underrepresented compared to non-stroke cases. This imbalance often results in biased machine learning models that perform poorly in predicting stroke occurrences. To address this issue, we evaluated three data-balancing methods: oversampling, undersampling, and the synthetic minority oversampling technique (SMOTE). Using a dataset comprising 42,617 healthy cases and 783 stroke cases, we assessed ten machine learning models, including logistic regression (LR), Naive Bayes (NB), support vector machine (SVM), decision tree (DT), random forest (RF), XGBoost, LightGBM, CatBoost, K-nearest neighbors (KNN), and artificial neural network (ANN). Our results indicate that combining SMOTE with the CatBoost model substantially enhances both prediction accuracy and sensitivity compared to using undersampling or oversampling alone. This underscores SMOTE’s effectiveness in addressing class imbalance in healthcare datasets. By highlighting the critical role of data balancing in improving prediction, this study offers practical insights for developing and clinically applicable predictive models.