Empirical Evaluation of an SVM-based Software Reliability Model

Support Vector Machines (SVMs) are known as some of the best learning models for pattern recognition, and an SVM can be used as a software reliability model to predict fault-prone modules from complexity metrics. We experimentally evaluated the prediction performance of an SVM model, comparing it with commonlyused conventional models including linear discriminant analysis, logistic regression, a classification tree, and a neural network. The results revealed that the SVM model exhibited showed the best performance among all the models tested.