Commit Classification Into Software Maintenance Activities: A Systematic Literature Review
Tjaša Heričko, Boštjan Šumak
Abstract
Commits represent an essential part of software development practices, serving as the means for collaboration and management of software changes made to a software project’s codebase. Changes are necessary for the survival of software, for instance, to fix faults, address security vulnerabilities, meet new functional requirements, and improve software performance. To aid software research and practice, several research endeavors attempted to automate the process of identifying the nature of the maintenance tasks performed in a commit based on the data associated with the commit. This paper presents a systematic literature review of supervised-learning-based models for commit classification into maintenance activities. Through the study selection process, 19 primary studies were identified, published between 2008 and 2023 in various journals and conference proceedings. For the independent variables for classification, features extracted from commit messages are prevalent, followed by features extracted from source code and commit metadata. In the majority of existing studies, multi-class classification approaches are used, whereas multi-label approaches are considered rarely. The most commonly used classification algorithms include Decision Tree, Random Forest, Naive Bayes, and Neural Network. Several datasets exist, consisting mainly of commits from different open-source projects based on the Java programming language. Some research gaps and open challenges were identified that can guide future research efforts.