Discriminant Subspace Alignment for Cross-project Defect Prediction

Cross-project defect prediction (CPDP) aims at recognizing defective software modules in a target project with the utilization of historical data from other source projects. Lately, CPDP has attracted much research interest. However, distribution discrepancy between the source and targert projects is known to have a negative effect on CPCP performance. Furthermore, most of the CPDP methods don't consider to explore the class label information in source data, thus their prediction performance may be limited. In the paper, we first introduce a new subspace alignment (SA) based domain adaptation method into CPDP, which can reduce the data distribution discrepancy between the source and target projects. Then, we propose a discriminant SA (DSA) approach for CPDP, the class label information of source project can be fully used. Experimental results from five public projects of NASA dataset demonstrate that DSA outperforms the related competing methods.