A model for software defect prediction using support vector machine based on CBA

Abstract

Software defection prediction is not only crucial for improving software quality, but also helpful for software test effort estimation. As is well-known, 80% of the fault happens in 20% of the modules. Therefore, we need to find out the most error prone modules accurately and correct them in time to save time, money, and energy. Support vector machine (SVM) is an advanced classification method that fits the defection classification. However, studies show that, the value of parameters of SVM model has a remarkable influence on its classification accuracy and the selection process lacks theory guidance that makes the SVM model uncertainty and low efficiency. In this paper, a CBA-SVM software defect prediction model is proposed, which take advantage of the non-linear computing ability of SVM model and optimisation capacity of bat algorithm with centroid strategy (CBA). Through the experimental comparison with other models, CBA-SVM is proved to have a higher accuracy.