Statistical model for prediction of software defect arrival pattern

Abstract

In spite of planning and adherence to standards and processes, defects are inevitable in any software development project. To meet high quality standards, software services firms need to have techniques in place for quality monitoring and control. The earlier the bugs are detected in the software lifecycle the less costly it is to fix them. To track project lifecycle, many off-the shelf tools are available in the market which facilitate project management, project scheduling, task and resource allocations, and defect management. Then there are specialized tools for each of the phases of the project. Microsoft Project is a popular project management tool, whereas HPs ALM (Application Lifecycle Management) helps track the various phases of a project and even provides Incident and Change Management features for managing software post production. Then there are specialized tools in the market solely for bug management and tracking like Bugzilla and JIRA. Almost all software services companies have been using one or more of these tools for years now and are sitting on a gold-mine of data which needs to be analyzed. Advancement in the field of statistics now enable these companies to have prediction models in place which can leverage data collected from past software projects to provide valuable insights on the quality of future software projects. An early prediction of the number of bugs that will surface later will help them make course corrections so that the product can be delivered at the committed date and with agreed quality. This project aims to explore few statistical models and apply the relevant one on a set of defect data collected from an on-going project in a large multinational company for a Banking and Financial Services customer. The company is operating at an SEI CMM Level 5 and is in the software business for many years.