Weighted SVMBoost based Hybrid Rule Extraction Methods for Software Defect Prediction

Abstract

The software testing efforts and costs are mitigated by appropriate automatic defect prediction models. So far, many automatic software defect prediction (SDP) models were developed using machine learning methods. However, it is difficult for the end users to comprehend the knowledge extracted from these models. Further, the SDP data is of unbalanced in nature, which hampers the model performance. To address these problems, this paper presents a hybrid weighted SVMBoost-based rule extraction model such as WSVMBoost and Decision Tree, WSVMBoost and Ripper, and WSVMBoost and Bayesian Network for SDP problems. The extraction of the rules from the opaque SVMBoost is carried out in two phases: (i) knowledge extraction, (ii) rule extraction. The experimental results on four NASA MDP datasets have shown that the WSVMBoost and Decision tree hybrid yielded better performance than the other hybrids and WSVM.