Application of feed-forward neural networks for software reliability prediction

Many analytical models have been proposed for modeling software reliability growth trends with different predictive capabilities at different phases of testing yet there still is a need to develop a model that can be applied for accurate predictions in a realistic environment. In this paper we describe a software reliability prediction model using feed-forward neural network for better reliability prediction through back-propagation algorithm and discuss the issues of network architecture and data representation methods. We demonstrate a comparative analysis between the proposed approach and three well known software reliability growth prediction models using seven different failure datasets collected from standard software projects to test the validity of the presented method. A numerical example also has been cited to illustrate the results that revealed significant improvement by using Artificial Neural Network (ANN) over conventional statistical models based on NHPP.