Software Defect Prediction: A Comparison Between Artificial Neural Network and Support Vector Machine

Abstract

Software industry has stipulated the need for good quality software projects to be delivered on time and within budget. Software defect prediction (SDP) has led to the application of machine learning algorithms for building defect classification models using software metrics and defect proneness as the independent and dependent variables, respectively. This work performs an empirical comparison of the two classification methods: support vector machine (SVM) and artificial neural network (ANN), both having the predictive capability to handle the complex nonlinear relationships between the software attributes and the software defect. Seven data sets from the PROMISE repository are used and the prediction models’ are assessed on the parameters of accuracy, recall, and specificity. The results show that SVM is better than ANN in terms of recall, while the later one performed well along the dimensions of accuracy and specificity. Therefore, it is concluded that it is necessary to determine the evaluation parameters according to the criticality of the project, and then decide upon the classification model to be applied.