Improving the performance of defect prediction based on evolution data

It is an undisputed fact that software continues to evolve. Software evolution is caused by requirement changes which often result in injection of defects. Existing defect prediction techniques mainly focus on utilizing the attributes of software work products, such as documents, source codes and test cases, to predict defects. Consider an evolving software as a species and its development process as a natural species' evolutionary process, the injection of defects may have the characters of a species and will be impacted by its evolution. A great many of researchers have studied the process of software evolution and proposed some evolution related metrics. In this study, a set of new metrics is first proposed based on evolutionary history to characterize software evolution process, and then a case study on building defect prediction models is presented. Experiments on six well-known open source projects achieved good performance, demonstrating the effectiveness of the proposed metrics. © Copyright 2016, Institute of Software, the Chinese Academy of Sciences. All rights reserved.