A Novel Feature Selection Approach based on Binary Particle Swarm Optimization and Ensemble Learning for Heterogeneous Defect Prediction

Abstract

Software defect prediction is an integral part of the software development process. Defect prediction helps focus on the grey areas beforehand, thus saving the considerable amount of money that is otherwise wasted in finding and fixing the faults once the software is already in production. One of the popular areas of defect prediction in recent years is Heterogeneous Defect Prediction, which predicts defects in a target project using a source project with different metrics. Through our paper, we provide a novel feature selection based approach, En-BPSO, based on binary particle swarm optimization, coupled with majority voting ensemble classifier based fitness function for heterogeneous defect prediction. The datasets we are using are MORPH and SOFTLAB. The results show that the En-BPSO method provides the highest Friedman mean rank amongst all the feature selection methods used for comparison. En-BPSO technique also helps us dynamically determine the optimal number of features to build an accurate heterogeneous defect prediction model.