Using Machine Learning Methods for Automatic Bug Assignment to Developers

Abstract

Background and Objectives: It is generally accepted that the highest cost in software development is associated with the software maintenance phase. In corrective maintenance, the main task is correcting the bugs found by the users. These bugs are submitted by the users to a Bug Tracking System (BTS). The bugs are evaluated by the bug triager and assigned to the developers to correct them. To find a related developer to correct the bug, recent developers’ activities and previous bug fixes must be examined. This paper presents an automated method to assign bugs to developers by identifying similarity between new bugs and previously reported bug reports.Methods: For automatic bug assignment, four clustering techniques (i.e. Expectation-Maximization (EM), Farthest First, Hierarchical Clustering, and Simple Kmeans) are used where a tag is created for each cluster that indicates an associated developer for bug correction. To evaluate the quality of the proposed methods, the clusters generated by the methods are compared with the labels suggested by an expert triager.Results: To evaluate the performance of the proposed method, we use real-world data of a large scale web-based system which is stored in the BTS of a software company. To select the appropriate algorithm for the clustering, the outputs of each clustering algorithm are compared to the labels suggested by the expert triager. The algorithm with closer output to the expert opinion is selected as the best algorithm. The results showed that EM and FarthestFirst clustering algorithms with 3% similarity error have the most similarity with the expert opinion.Conclusion: the results obtained by the algorithms show that we can successfully apply them for bug assignment in real-world software development environments.