Scholarlink

<p>Software defect prediction that will enable us to create high-quality software at lower costs as well as reduce development costs. However, traditional prediction methods usually do not provide the required precision for effective defect management. Early detection of error-prone modules allows software projects leaders to set priorities in testing, concentrating their attention on the most probable modules. The current paper introduces a hybrid technique consisting of K-member Fuzzy Clustering and a Cost-sensitive Multi-objective Metaheuristic-driven Ensemble Feature Learning (also called KFC-CMMEFL), which is a new method of error detection in software. The suggested technique starts with the data preparation phase, during which K-member fuzzy clustering and oversampling techniques are applied in order to handle the missing data problem and evaluate redundant features. Next, the offline optimization phase is deployed where the NSGA-II multi-objective algorithm adjusts the. model’s hyperparameters, including classifier-specific features, ensemble weights, final classification threshold, and other controllable parameters. After being optimized with 10-fold cross-validation, the optimized KFC-CMMEFL model then changes over to an online phase which leads to even better defect classification. Through the use of balanced fuzzy clustering and a cost-sensitive metaheuristic-optimized ensemble feature learning model, our approach is able to achieve a better trade-off between true-positives and false-positives, thus, the accuracy of defect predictions climbs. The experiments we conducted on 10 different datasets from tera-PROMISE showed that KFC-CMMEFL outperforms the referent methods in software defect prediction field.</p>

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

Efficient software defect prediction using fuzzy K-member clustering and metaheuristic-driven ensemble feature learning model

  • 期刊论文

  • 发表日期：

    2026年2月12日

查看全文
  • Shima Javadimoghadam,
  • Seyed Mojtaba Sabagh-Jafari,
  • Amid Khatibi Bardsiri

摘要

Software defect prediction that will enable us to create high-quality software at lower costs as well as reduce development costs. However, traditional prediction methods usually do not provide the required precision for effective defect management. Early detection of error-prone modules allows software projects leaders to set priorities in testing, concentrating their attention on the most probable modules. The current paper introduces a hybrid technique consisting of K-member Fuzzy Clustering and a Cost-sensitive Multi-objective Metaheuristic-driven Ensemble Feature Learning (also called KFC-CMMEFL), which is a new method of error detection in software. The suggested technique starts with the data preparation phase, during which K-member fuzzy clustering and oversampling techniques are applied in order to handle the missing data problem and evaluate redundant features. Next, the offline optimization phase is deployed where the NSGA-II multi-objective algorithm adjusts the. model’s hyperparameters, including classifier-specific features, ensemble weights, final classification threshold, and other controllable parameters. After being optimized with 10-fold cross-validation, the optimized KFC-CMMEFL model then changes over to an online phase which leads to even better defect classification. Through the use of balanced fuzzy clustering and a cost-sensitive metaheuristic-optimized ensemble feature learning model, our approach is able to achieve a better trade-off between true-positives and false-positives, thus, the accuracy of defect predictions climbs. The experiments we conducted on 10 different datasets from tera-PROMISE showed that KFC-CMMEFL outperforms the referent methods in software defect prediction field.