The Cold-start Problem in Software Fault Prediction

Download or read book The Cold-start Problem in Software Fault Prediction written by Inbal Roshanski and published by . This book was released on 2020 with total page pages. Available in PDF, EPUB and Kindle. Book excerpt: Software is an integral part of our lives today. Unfortunately, the more sophisticated and complicated software becomes, the greater the chance of failures. Predicting the probability of software components being faulty can help maintaining the software effectiveness. A key factor to the success of prediction algorithms is the amount and quality of historical data of the project collected by the version control and issue tracker tools. However, for new projects, for example, there is no historical data to learn from. This is known as the cold-start problem. Previous work proposed cross-project software fault prediction models, where fault prediction models of other projects are used to determine whether new project's components are faulty or not. In this paper we suggest a novel component-sensitive cross-project software fault prediction approach (OSCAR). OSCAR proceeds in two steps. First, it separately classifies each component in the new project to its most similar project among a set of other projects. Then, OSCAR uses the fault prediction model of that project to predict whether the component in the new project is faulty. This approach is in contrast to previous work that try to find one suitable model for all the components in the new project. Furthermore, we suggest an improvement to OSCAR, by using clustering algorithm combined with it. Evaluation, conducted on three datasets which includes 43 software projects, shows that the prediction of OSCAR is more accurate than state-of-the-art competitive algorithms.