Software metrics reduction for fault-proneness prediction of software modules

Abstract

It would be valuable to use metrics to identify the fault-proneness of software modules. However, few research works are on how to select appropriate metrics for fault-proneness prediction currently. We conduct a large-scale comparative experiment of nine different software metrics reduction methods over eleven public-domain data sets from the NASA metrics data repository. The Naive Bayes data miner, with a log-filtering preprocessor on the numeric data, is utilized to construct the prediction model. Comparisons are based on the analysis of variance. Our conclusion is that, reduction methods of software metrics are important to build adaptable and robust software fault-proneness prediction models. Given our results on Naive Bayes and log-filtering, discrete wavelet transformation outperforms other reduction methods, and correlation-based feature selection with genetic search algorithm and information gain can also obtain better predicted performance. © 2010 Springer-Verlag.