Scholarlink

<p>Software development greatly benefits from defect prediction, which will prioritize testing modules with the highest potential for defects. Traditional effort-aware defect prediction (EADP) methods usually contain regression or classification methods, which might not fully represent the complexity of software features. To improve feature selection and ranking, this study offers an improved EADP model that uses modern machine learning algorithms such as XGBoost, AdaBoost, and Random Forest. This study integrates a synergy-based feature fusion and attention mechanism to improve the prioritization of defective modules. The suggested method improves prediction efficiency and accuracy by directly adjusting the Proportion of Found Bugs at 20% of LOC (PofB@20%). The findings from experiments gathered across different public datasets show that the proposed model improves current methods with considerable improvements in PofB at 20% and decreases in initial false alarms (IFA). These findings show that combining gradient-boosting approaches with synergy-based fusion and attention-weighting tricks can significantly improve EADP models, bringing significant benefits for software testing and quality control.</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

Improving effort-aware defect prediction using machine learning methods

  • 期刊论文

  • 发表日期：

    2026年6月29日

查看全文
  • Alireza Mahdi Barzi,
  • Amirfarhad Farhadi,
  • Azadeh Zamanifar

摘要

Software development greatly benefits from defect prediction, which will prioritize testing modules with the highest potential for defects. Traditional effort-aware defect prediction (EADP) methods usually contain regression or classification methods, which might not fully represent the complexity of software features. To improve feature selection and ranking, this study offers an improved EADP model that uses modern machine learning algorithms such as XGBoost, AdaBoost, and Random Forest. This study integrates a synergy-based feature fusion and attention mechanism to improve the prioritization of defective modules. The suggested method improves prediction efficiency and accuracy by directly adjusting the Proportion of Found Bugs at 20% of LOC (PofB@20%). The findings from experiments gathered across different public datasets show that the proposed model improves current methods with considerable improvements in PofB at 20% and decreases in initial false alarms (IFA). These findings show that combining gradient-boosting approaches with synergy-based fusion and attention-weighting tricks can significantly improve EADP models, bringing significant benefits for software testing and quality control.