Cross-Project Software Defect Prediction with Subdomain Adaptation

In recent years, Cross-Project Software Defect Prediction (CPDP) has gained prominence in the field of software engineering. The performance of CPDP is impacted by the variations in feature distributions between source and target projects. Many researches have proposed different method to address this challenge. However, previous studies that employ Domain Adaptation to align the distribution rarely consider the subdomain. To overcome this constraint, we proposed a novel approach called Deep Adversarial Subdomain Adaptation for CPDP. The method consists of four parts: a feature extractor for extracting features from the source and target projects, a classifier for predict the defects, a domain discriminator for predicting domain label and a LMMD module for aligning the subdomain distribution. Our experiments were using AEEEM datasets and the outcomes demonstrate that our method achieves a better performance.