Scholarlink

This study presents a machine learning framework for predicting survival outcomes in eye cancer patients using a real-world clinical dataset. Predicting the survival outcomes help us in personalizing treatment plans and improve patient care. It uses six different feature selection techniques such as RST, ZST, ZSTR, CCRA [6], PCA, GA are used along with the Synthetic Minority Over-sampling Technique (SMOTE) to address the class imbalance. Sixteen classifiers which include Naive Bayes variants, SVMs, Decision Trees, ensemble models, and others, were evaluated using 5-fold cross-validation. Results show that highest AUC of 0.8606 is achieved while using PCA+SMOTE with ExtraTreesClassifier, while the best F1-Score of 0.7999 is achieved with CCRA+SMOTE using ExtraTreesClassifier. Bagging with DecisionTree on SMOTE-processed original features yields the highest accuracy of 82.54%. The findings underscore the value of combining data balancing with feature selection for robust predictive modeling in imbalanced clinical datasets.

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

Predicting Eye Cancer Survival Using Feature Selection and Ensemble Machine Learning Techniques

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • B. Lakshmi Nivas,
  • Lov Kumar,
  • D. V. N. Siva Kumar,
  • Vikram Singh

摘要

This study presents a machine learning framework for predicting survival outcomes in eye cancer patients using a real-world clinical dataset. Predicting the survival outcomes help us in personalizing treatment plans and improve patient care. It uses six different feature selection techniques such as RST, ZST, ZSTR, CCRA [6], PCA, GA are used along with the Synthetic Minority Over-sampling Technique (SMOTE) to address the class imbalance. Sixteen classifiers which include Naive Bayes variants, SVMs, Decision Trees, ensemble models, and others, were evaluated using 5-fold cross-validation. Results show that highest AUC of 0.8606 is achieved while using PCA+SMOTE with ExtraTreesClassifier, while the best F1-Score of 0.7999 is achieved with CCRA+SMOTE using ExtraTreesClassifier. Bagging with DecisionTree on SMOTE-processed original features yields the highest accuracy of 82.54%. The findings underscore the value of combining data balancing with feature selection for robust predictive modeling in imbalanced clinical datasets.