• Brahmaleen Kaur Sidhu

Software refactoring is essential for maintaining design quality, yet automated support for refactoring UML class diagrams remains largely rule‑ or search‑based and does not learn reusable strategies. This paper proposes a reinforcement learning (RL) framework for automated UML class‑diagram refactoring that learns refactoring policies from interaction with design models. We formalise refactoring as a Markov Decision Process in which states represent UML models, actions are parameterised semantics‑preserving refactorings (e.g., Move Method, Extract Class), and rewards are based on changes in a composite design‑quality function derived from established object‑oriented metrics. A UML refactoring engine built on EMF/UML2 is exposed via a REST API and wrapped as a Gym‑compatible environment, on which a Proximal Policy Optimization (PPO) agent is trained to maximise expected cumulative quality improvement. Experiments on UML class diagrams reverse‑engineered from real Java systems show that the learned policy achieves larger improvements in metric‑based design quality than random refactoring and a greedy hill‑climbing heuristic, and is competitive with a simple genetic search baseline. These results indicate that RL is a promising approach for learning reusable, quality‑driven refactoring strategies at the UML model level.

Design Quality Metrics, Reinforcement Learning, Search-Based Software Engineering, Software Refactoring, UML Class Diagrams

"Reinforcement Learning-Based Framework for Refactoring of UML Class Diagrams", International Journal of Emerging Technologies and Innovative Research (www.jetir.org), ISSN:2349-5162, Vol.13, Issue 1, page no.a442-a450, January-2026, Available :http://www.jetir.org/papers/JETIR2601054.pdf

"Reinforcement Learning-Based Framework for Refactoring of UML Class Diagrams", International Journal of Emerging Technologies and Innovative Research (www.jetir.org | UGC and issn Approved), ISSN:2349-5162, Vol.13, Issue 1, page no. ppa442-a450, January-2026, Available at : http://www.jetir.org/papers/JETIR2601054.pdf