Scholarlink

This study focuses on optimizing the K-Nearest Neighbors (KNN) algorithm to predict hematomas associated with the Totally Implantable Venous Access Ports system (TIVAPS). Introduced in oncology in the 1980s, TIVAPS provide a durable solution for patients requiring long-term treatments but pose risks of hematomas, leading to severe complications. Using a retrospective dataset of 1084 patients from the Hassan II Oncology Centre in OUJDA, we developed predictive models by applying KNN with ten-fold cross-validation. To improve the model’s performance, we employed GridSearchCV for hyperparameters optimization, adjusted the decision threshold, and combined KNN with logistic regression. The results indicate that optimization via GridSearchCV achieved an accuracy of 87%, precision of 84%, recall of 66%, and an F1-score of 74%. Threshold adjustment improved the Precision-Recall balance with a recall of 71%, while the combination with logistic regression (LR) maintained robust performance. These results demonstrate the effectiveness of advanced machine learning techniques in improving patient care by anticipating and managing hematoma risks, leading to better patient outcomes and optimized healthcare resources.

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

Optimization of K-Nearest Neighbors for Hematoma Prediction in Totally Implantable Venous Access Ports

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Hanane El Oualy,
  • Bekkay Hajji,
  • Mouhsine Omari,
  • Hamid Madani

摘要

This study focuses on optimizing the K-Nearest Neighbors (KNN) algorithm to predict hematomas associated with the Totally Implantable Venous Access Ports system (TIVAPS). Introduced in oncology in the 1980s, TIVAPS provide a durable solution for patients requiring long-term treatments but pose risks of hematomas, leading to severe complications. Using a retrospective dataset of 1084 patients from the Hassan II Oncology Centre in OUJDA, we developed predictive models by applying KNN with ten-fold cross-validation. To improve the model’s performance, we employed GridSearchCV for hyperparameters optimization, adjusted the decision threshold, and combined KNN with logistic regression. The results indicate that optimization via GridSearchCV achieved an accuracy of 87%, precision of 84%, recall of 66%, and an F1-score of 74%. Threshold adjustment improved the Precision-Recall balance with a recall of 71%, while the combination with logistic regression (LR) maintained robust performance. These results demonstrate the effectiveness of advanced machine learning techniques in improving patient care by anticipating and managing hematoma risks, leading to better patient outcomes and optimized healthcare resources.