Scholarlink

Software testing is an essential phase in the development lifecycle, as defects introduced during coding can escalate in complexity if detected at later stages. Identifying these defects early reduces risks and ensures optimal use of testing resources. Software Defect Prediction (SDP) is a method used to classify software modules as either defective or non-defective, with various techniques proposed to enhance automation and accuracy in defect detection. This study presents a novel SDP approach utilizing Independent Component Analysis (ICA) for feature selection. The proposed method is evaluated across multiple datasets, including JM1, MC2, MW1, and PC3, achieving accuracy rates of 97.18%, 98.07%, 98.61%, and 98.43%, respectively. Different classifiers, such as Random Forest (RF), AdaBoost (AB), Decision Tree (DT), and Naïve Bayes (NB), are used to assess predictive performance. Among them, the RF classifier, when integrated with ICA, delivers the highest accuracy of 98.61% on the MW1 dataset, employing six-fold cross-validation and 23 selected features. The findings highlight that ICA-based feature selection enhances SDP model performance, providing an effective approach to improving software quality and reliability.

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

An Automated Software Defect Prediction Model Using Machine Learning Approaches

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Jayanta Kumar Mishra,
  • Debasish Pradhan,
  • Chitta Ranjan Sahoo,
  • Rednam S. S. Jyothi,
  • Lopamudra Das

摘要

Software testing is an essential phase in the development lifecycle, as defects introduced during coding can escalate in complexity if detected at later stages. Identifying these defects early reduces risks and ensures optimal use of testing resources. Software Defect Prediction (SDP) is a method used to classify software modules as either defective or non-defective, with various techniques proposed to enhance automation and accuracy in defect detection. This study presents a novel SDP approach utilizing Independent Component Analysis (ICA) for feature selection. The proposed method is evaluated across multiple datasets, including JM1, MC2, MW1, and PC3, achieving accuracy rates of 97.18%, 98.07%, 98.61%, and 98.43%, respectively. Different classifiers, such as Random Forest (RF), AdaBoost (AB), Decision Tree (DT), and Naïve Bayes (NB), are used to assess predictive performance. Among them, the RF classifier, when integrated with ICA, delivers the highest accuracy of 98.61% on the MW1 dataset, employing six-fold cross-validation and 23 selected features. The findings highlight that ICA-based feature selection enhances SDP model performance, providing an effective approach to improving software quality and reliability.