期刊名称：Journal of Software Engineering and Applications
印刷版ISSN：1945-3116
电子版ISSN：1945-3124
出版年度：2017
卷号：10
期号：03
页码：288-309
DOI：10.4236/jsea.2017.103017
语种：English
出版社：Scientific Research Publishing
摘要：Software fault prediction is one of the most fundamental but significant management techniques in software dependability assessment. In this paper we concern the software fault prediction using a multilayer-perceptron neural network, where the underlying software fault count data are transformed to the Gaussian data, by means of the well-known Box-Cox power transformation. More specially, we investigate the long-term behavior of software fault counts by the neural network, and perform the multi-stage look ahead prediction of the cumulative number of software faults detected in the future software testing. In numerical examples with two actual software fault data sets, we compare our neural network approach with the existing software reliability growth models based on nonhomogeneous Poisson process, in terms of predictive performance with average relative error, and show that the data transformation employed in this paper leads to an improvement in prediction accuracy.
关键词：Software Reliability;Artificial Neural Network;Box-Cox Power Transformation;Long-Term Prediction;Fault Count Data;Empirical Validation
