Efficient prediction of software fault proneness modules using support vector machines and probabilistic neural networks

Abstract

A software fault is a defect that causes software failure in an executable product. Fault prediction models usually aim to predict either the probability or the density of faults that the code units contain. Many fault prediction models using software metrics have been proposed in the Software Engineering literature. This study focuses on evaluating high-performance fault predictors based on support vector machines (SVMs) and probabilistic neural networks (PNNs). Five public NASA datasets from the PROMISE repository are used to make these predictive models repeatable, refutable, and verifiable. According to the obtained results, the probabilistic neural networks generally provide the best prediction performance for most of the datasets in terms of the accuracy rate.