Applying Neural Network Approach with Imperialist Competitive Algorithm for Software Reliability Prediction

Software systems exist in different critical domains. Software reliability assessment has become a critical issue due to the variety levels of software complexity. Software reliability, as a sub-branch of software quality, has been exploited to evaluate to what extend the desired software is trustable. To overcome the problem of dependency to human power and time limitation for software reliability prediction, researchers consider soft computing approaches such as Neural Network and Fuzzy Logic. These techniques suffer from some limitations including lack of analyzing mathematical foundations, local minima trapping and convergence problem. This study develops a novel model for software reliability prediction through the combination of Multi-Layer Perceptron Neural Network (MLP) and Imperialist Competitive Algorithm (ICA). The proposed model has solved some of the problems of existing methods such as convergence problem and demanding on huge number of data. This model can be used in complicated software systems. The results prove that both training and testing phases of this model outperform existing approaches in terms of predicting the number of software failures.