A tool suite for estimation and prediction of software dynamic defect models

Abstract

As a common software engineering practice, software dynamic defect models are used to estimate and predict the software testing process progress, effectiveness, and the number of future defects over the next weeks. Practitioners use these dynamic defect models to ensure that the delivery of software to customers is possible from the quality point of view and to predict the release date. Old literature suggested several classic defect models including Putnam, Exponential, Rayleigh and Weibull. Recent literature claimed that modern projects follow linear combinations of Rayleigh due to projects complexity. This claim verification has not been generalized because the project samples size was very small. This paper proposes a tool suite for dynamic defect models. The tool suite consists of an open repository of dynamic defects empirical data and many supporting tools. Data concerning defects are collected from several software projects and products and added to the open repository. This includes open source software and commercial software projects. The proposed tools are designed and implemented and made publicly available on the web. They are used to view the dynamic defects, find the best dynamic defect model that fits the data according to several performance criteria and predict future number of defects. The application of these tools on the empirical data showed that linear combinations of Rayleigh and Weibull has better performance than classic models in both curve fitting and predictability of commercial software.