Scholarlink

This paper proposes an integrated framework for predicting class-level restructuring opportunities using meta-learning techniques. The proposed framework uses stacking, boosting, and short-term learning algorithms, supported by a hybrid feature selection approach that combines the SoA and Spider Monkey optimization (SMO) algorithms. The system is demonstrated using four open-source Java datasets from projects. After data cleaning and feature selection, grid search was used to fine-tune and train the models. The hybrid feature selection approach significantly improved the prediction performance of the models. On most datasets, SVC, XGBoost, CatBoost, and RNM achieved a maximum possible accuracy of 1.00. The accuracy, specificity, recall, F1 score, and cross-entropy error measure demonstrated the effectiveness of the proposed system during evaluation.

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

Improving the Refactoring Prediction at Class-Level Using Meta-learning approach

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Shahbaa I. Khaleel,
  • Rasha Ahmed Mahmood

摘要

This paper proposes an integrated framework for predicting class-level restructuring opportunities using meta-learning techniques. The proposed framework uses stacking, boosting, and short-term learning algorithms, supported by a hybrid feature selection approach that combines the SoA and Spider Monkey optimization (SMO) algorithms. The system is demonstrated using four open-source Java datasets from projects. After data cleaning and feature selection, grid search was used to fine-tune and train the models. The hybrid feature selection approach significantly improved the prediction performance of the models. On most datasets, SVC, XGBoost, CatBoost, and RNM achieved a maximum possible accuracy of 1.00. The accuracy, specificity, recall, F1 score, and cross-entropy error measure demonstrated the effectiveness of the proposed system during evaluation.