Predicting bug inducing source code change patterns

Abstract

A change in source code without the prior analysis of its impact may generate one or more defects. Fixing of such defects consumes maintenance time which ultimately increases the cost of software maintenance. Therefore, in the recent years, several research works have been done to develop techniques for the automatic impact analysis of changes in source code. In this paper, we propose to use Frequent Pattern Mining (FPM) technique of machine learning for the automatic impact analysis of those changes in source code which may induce bugs. Therefore, to find patterns associated with some specific types of software changes, we applied FPM's algorithms' Apriori and Predictive Apriori on the stored data of software changes of the following three Open-Source Software (OSS) projects: Mozilla, GNOME, and Eclipse. We grouped the data of software changes into two major categories: changes to meet bug fixing requirements and changes to meet requirements other than bug fixing. In the case of bug fixing requirements, we predict source files which are frequently changed together to fix any one of the following four types of bugs related to: memory (MEMORY), variables locking (LOCK), system (SYSTEM) and graphical user interface (UI). Our experimental results predict several interesting software change patterns which may induce bugs. The obtained bug inducing patterns have high confidence and accuracy value i.e., more than 90%.