<![CDATA[Software defect prediction (SDP) plays a crucial role in improving software quality by enabling the early detection of faulty modules during the development phase. However, class imbalance within software defect datasets remains a significant challenge that adversely impacts prediction accuracy. This study aims to address this issue by implementing ensemble learning methods—specifically Bagging and AdaBoost—combined with the C4.5 decision tree algorithm to enhance classification performance. The research utilized five well-known datasets from the NASA MDP Repository (CM1, JM1, KC1, KC2, and PC1), each containing comprehensive software metrics and defect labels. The methodology involved several stages: data preprocessing (normalization and discretization), model training using 10-fold cross-validation, and performance evaluation through metrics such as accuracy and Area Under the Curve (AUC). Results indicate that both ensemble methods outperformed the standalone C4.5 algorithm across all datasets. Notably, the AdaBoost + C4.5 model yielded the highest accuracy in most scenarios, with the PC1 dataset reaching 97.20% accuracy. In comparison, C4.5 alone and C4.5 with Bagging recorded lower values, demonstrating the significant impact of adaptive weighting in AdaBoost. These findings affirm that ensemble learning, particularly AdaBoost, effectively mitigates the impact of class imbalance and improves prediction performance in SDP tasks.]]>
Copyrights © 2025
