Prediction Interval of Cumulative Number of Software Faults Using Multilayer Perceptron

Software reliability is one of the most important attributes in software quality metrics. To predict the number of software faults detected in testing phase, many approaches have been applied during the last four decades. Among them, the neural network approach plays a significant role to estimate and predict the number of software fault counts. In this paper, we focus on a prediction problem with the common multilayer perceptron neural networks, and derive the predictive interval of the cumulative number of software faults in sequential software testing. We apply the well-known back propagation algorithm for feed forward neural network architectures and the delta method to construct the prediction intervals. In numerical experiments with four real software development project data sets, we evaluate the one-stage look ahead prediction interval of the cumulative number of software faults, and compare three data transform methods, which are needed for pre-processing the underlying data, in terms of average relative error, coverage rate and predictive interval width.