EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Maintainability of LLM-Generated Code and Student’s Understanding

Md. Mehedi Hasan, Md Forhad Rabbi, Fazle Rabbi Rakib

SUST Journal of Science and Technology (SUST JST) · 08.11.2026 · 35 (2)

DOI: 10.63512/sustjst.2025.2003 ↗

Kurz gefasst

Fachartikel in der Zeitschrift SUST Journal of Science and Technology (SUST JST), online erschienen am 08.11.2026. Verfasst von Md. Mehedi Hasan und 2 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Large Language Models (LLMs) are frequently used by students for programming assignments and capstone projects, yet their impacts on software maintainability and students’ ability to explain and maintain their own codebases remain unclear in authentic educational settings. This study examines maintainability and code understanding as jointly examined outcomes in an undergraduate Web Technology course. We analyze 48 student GitHub repositories (3,043 commits) using commit-level SonarQube static analysis to track maintainability and risk-relevant indicators, including code smells, SQALE technical-debt measures, duplication, complexity, and security/reliability ratings. We complement repository-level measurements with student explainability (understanding) scores on a 0–10 scale (n = 104), reflecting students’ ability to explain code segments from their projects. Across repositories, 5/48 (10.4%) failed the SonarQube quality gate. Security and reliability concerns were concentrated in a subset of projects. Our investigation exhibits 9/48 (18.75%) blocker-level security status, and 2/48 (4.17%) high security issues. Whereas reliability issues included 2/48 (4.17%) with blocker-level reliability and 6/48 (12.5%) with high reliability issues, while 32/48 (66.67%) fell in the medium range. In addition, security hotspots requiring review were prevalent (42 repositories). On the learning side, 33% of students scored ≤ 5/10, indicating limited ability to explain their code, and approximately 2% could not provide a meaningful explanation. Overall, the results suggest that a notable fraction of student projects retain maintainability and security/reliability risks, and that a non-trivial subset of students demonstrate low explainability of their own code. These findings motivate instructional scaffolding and assessment practices that emphasize code review, refactoring, and explanation alongside functional correctness, especially in courses where LLM-based tools are available.

Zitieren

APA

Hasan, M. M., Rabbi, M. F., & Rakib, F. R. (2026). Maintainability of LLM-Generated Code and Student’s Understanding. SUST Journal of Science and Technology (SUST JST), 35(2). https://doi.org/10.63512/sustjst.2025.2003

Vancouver

Hasan MM, Rabbi MF, Rakib FR. Maintainability of LLM-Generated Code and Student’s Understanding. SUST Journal of Science and Technology (SUST JST). 2026;35(2). doi:10.63512/sustjst.2025.2003

BibTeX
@article{Hasan2026,
  author = {Hasan, Md. Mehedi and Rabbi, Md Forhad and Rakib, Fazle Rabbi},
  title = {Maintainability of LLM-Generated Code and Student’s Understanding},
  journal = {SUST Journal of Science and Technology (SUST JST)},
  year = {2026},
  volume = {35},
  number = {2},
  doi = {10.63512/sustjst.2025.2003}
}
Kostenlos starten →