Collective transfer learning for defect prediction

Abstract

Most software defect prediction approaches require extensive data from the project under test for training. However, for a new project, enough training data is often not available. It is therefore necessary to build a predictive model using the data from other relevant projects and then use the model to predict defects in the target project. Such direct cross-project prediction performance can still be improved, mainly due to the distribution differences between the source and target projects, as well as the uncertainty in determining which source project should be selected to train the model. In this work, we propose a collective training mechanism for defect prediction (CTDP), which includes two phases: source data expansion phase and adaptive weighting phase. CTDP makes the feature distributions of source and target projects similar to each other by transfer learning, and uses the particle swarm optimization algorithm to comprehensively consider the multiple source projects to predict the target project. The experiments on a total of 28 projects in 5 groups show that our method can improve the performance of cross-project defect prediction, achieving the sate-of-the-art results.