Prediction of Software Readiness Using Neural Network

In this paper, we explore the behaviour of neuralnetwork in predicting software readiness. Our neural networkmodel aims to predict the number of faults (including objectoriented faults) of a software under development. We use Wardneural network that is a backpropagation network with differentactivation functions. Different activation functions are applied tohidden layer slabs to detect different features in a patternprocessed through a network. In our experiments, hyperbolictangent, Gaussian, Gaussian-complement and linear functions areused as activation functions to improve prediction. This paperalso compares the prediction results from multiple regressionmodel and neural network model. Object-oriented design metricsare used as the independent variables in our study. Our study isconducted on three industrial real-time systems that contain anumber of natural faults that has been reported over a period ofthree years