Ensemble and Cross-Project Software Reliability Growth Models for Safety-Critical Systems

Abstract

Software Reliability Growth Models (SRGMs) have been widely used for software reliability estimation, measurement of the current state, and prediction of the future state, based on results of software testing activities. Although there are various models that have been introduced to predict software reliability, none of them provides a common or general solution. In this study, we propose a new ensemble model (EM), which helps to prevent any extra work while finding out an SRGM, where the project data is fitting and suitable mostly. This proposed model tries to approach the SRGM with the lowest training error and also handles models with slightly worse training but good test results. The EM we created with this motivation has achieved its purpose. It has substantially saved us from the uncertainty of the correct SRGM selection process and helps us to get satisfying results in many cases. The other approach is the ensemble of cross projects (ECP), with this method, much better results are obtained by combining the results of a single cross-project prediction on ongoing projects. Good results are obtained through similarity analysis and proportioning according to the number of errors per unit time before combining. With the help of this study, we believe that any practitioner or researcher, who is interested in SRGMs will benefit from our results.