Source Project Selection for Cross-Project Software Defect Prediction using Clustering Approach

Software Defect prediction (SDP) has an effective role in improving the software system quality. However, defect prediction models using local data have poor performance as the data available is inadequate. Multi-Source Cross-Project software defect prediction (MCPDP) uses the knowledge gained from source projects to detect the defects in the target project. In this paper, a novel clustering approach is proposed to select source projects for a given target project. The proposed approach is compared with existing source project selection strategies, such as Random selection, selection using Recommendation scores and using all available projects as source projects along with different classifiers, namely Naive Bayes(NB), AdaBoost and TSBoostDF. The performance evaluation is done using F1-score, G-mean, MCC and Balance. Wilcoxon Signed-Rank (WSR) hypothesis test is conducted to statistically compare the models. The statistical comparison shows that the proposed approach outperforms the existing source project selection strategies.