Software Defect Prediction Using Support Vector Machine

Abstract

Software defect prediction is important task during software development Lifecycle as it can help managers to identify the most defect-proneness modules. Thus, it can reduce the test cost and assign testing resources efficiently. To make sure if the software is defective or not, there are many classification methods that can be used such as Decision Tree, Recognition, Support Vector Machine, Neural Network (ANN), Naive Bayes. However, Support Vector Machine (SVM) has not been used extensively for such problem, because of its instability when applied on different datasets and parameter settings. The main parameter that influences the accuracy is the choice of kernel function. Therefore, this research attempts to study the performance of SVM with different kernels. The use of kernel functions has not been studied thoroughly in previous papers. Various public datasets from PROMISE project are used to empirically validated our hypothesis. The results demonstrate that there is no kernel function that can give stable performance across different experimental settings. In addition, the use of PCA as feature reduction algorithm shows slight accuracy improvement over some datasets.