Design-Level Refactoring Using Genetic Algorithms in an Academic System

Abstract

Legacy software systems undergo structural degradation due to continuous evolution, which makes maintenance increasingly complex. This study investigates the effectiveness of design-level refactoring using a genetic algorithm (GA)-based class decomposition method in a legacy academic information system of FILKOM. We utilize static analysis via CodeMR to detect code smells and evaluate maintainability using the Maintainability Index (MI). Our findings reveal a significant increase in median MI from 28.25 to 70.90 post-refactoring. While formal statistical significance was not reached (p = 0.10), the effect size was consistently strong (r = 1.00), confirming the positive impact of the proposed approach. The study contributes a practical, replicable method for improving maintainability in real-world systems, supported by semantic similarity and usability-based decomposition. This work highlights the potential of design-level optimization using evolutionary algorithms and calls for further multi-domain validation.