Effective software defect prediction using support vector machines (SVMs)

Software defect prediction (SDP) plays a key role in the timely delivery of good quality software product. In the early development phases, it predicts the error-prone modules which can cause heavy damage or even failure of software in the future. Hence, it allows the targeted testing of these faulty modules and reduces the total development cost of the software ensuring the high quality of end-product. Support vector machines (SVMs) are extensively being used for SDP. The condition of unequal count of faulty and non-faulty modules in the dataset is an obstruction to accuracy of SVMs. In this work, a novel filtering technique (FILTER) is proposed for effective defect prediction using SVMs. Support vector machine (SVM) based classifiers (linear, polynomial and radial basis function) are designed utilizing the proposed filtering technique over five datasets and their performances are evaluated. The proposed FILTER enhances the performance of SVM based SDP model by 16.73%, 16.80% and 7.65% in terms of accuracy, AUC and F-measure respectively.