Title: Severity assessment of code smells using multi-criteria approach

Authors: Abhishilpa Nandini; Randeep Singh; Amit Rathee

Addresses: Department of Computer Science and Engineering, IEC University, Baddi, Solan, HP, 174103, India ' Department of Computer Science and Engineering, IEC University, Baddi, Solan, HP, 174103, India ' Department of Computer Science, Government College, Barota, Sonipat, HR, 131301, India

Abstract: Context: Code smells indicate that the source code is flawed and has design/structural problems. Over time, it causes reduced maintainability, understandability, and other quality issues. Existing tools and techniques generally return a large number of potential code smells making it difficult to handle all of them because of time and/or maintenance budget constraints. Thus, prioritisation or severity classification is a viable solution that reduces code smell volume by filtering critical code smells. Objective: Previous research pays very little attention to the direction of code smell severity classification and they mainly use trivial software metrics. Many other important aspects such as change history, co-occurrence patterns, modular importance, developer's perception, and severity are simply ignored. Therefore, this paper targets exploring new factors that help in improving severity classification. Method: We introduce a new multi-criteria-based severity classification approach that combines four key factors viz co-occurrence pattern among code smells, severity, module importance, and activeness of code smells to improve the quality of code smell prioritisation and severity classification. We proposed three metrics. The proposed approach is empirically evaluated on state-of-the-art open-source software systems. Results: During the empirical evaluation, about 66% of classes are found to be affected by code smells. Moreover, a sufficiently high number of classes are found in medium, high, and very high severity (VHS) categories. Further, we find that the proposed approach significantly improves severity classification results. Finally, the proposed approach is found to return justifiable accuracy that ranges between 78-93%. Conclusions: Based on the obtained results, the proposed approach is found to be useful in estimating severity and obtaining potential critical candidates for refactoring.

Keywords: code smell; prioritisation; ranking; severity classification; modular importance; cooccurrence; code smell activeness.

DOI: 10.1504/IJSSE.2026.156327

International Journal of System of Systems Engineering, 2026 Vol.16 No.4, pp.450 - 473

Received: 13 Mar 2024
Accepted: 12 May 2024
Published online: 13 Sep 2026 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article