Software reliability reckoning by applying neural network algorithm

Software reliability is the outlook of fault-free activity of software for a specified period in a specified environment. Many researches are conducted for increasing the software reliability. The processes in order to increase the software reliability can be determined in three steps such as-software representation, software evaluation and software amendment. Each of these processes is very much essential in order to improve the reliability of the software. The software is one of the most vital parts of many commercial, industrial and military operations. As the software is now being applied in many safety-critical systems, therefore it has become a significant research area. In order to assess the software engineering technologies, software reliability measure is used. Many metrics are proposed for enhancing the software reliability. Machine learning approaches are the most appropriate ways for evaluating several bounds of software reliability. In this research paper, the authors have implemented 14 training algorithms accessible in NNtool box in MATLAB on a cost-effectively accessible dataset which is UIMS. The training algorithms present in NNtool box creates, trains and simulates the network. Moreover, the performance was estimated based on R2 value. Nevertheless, the experimental outcomes demonstrated that TRAINBR algorithm was found to be the best amidst all the training algorithms present in NNtool box. Furthermore, the results denote that neural network technique can be efficiently used for reliability assessment.