Multi-Dimensional Genetic Algorithm-Based Remodularization Integrating Structural, Semantic, and Behavioral Dependencies

Architectural erosion in software systems is inevitable and it results in degraded modular quality that directly causes increased maintenance efforts along with reduced comprehensibility for the underlying software system. Software remodularization is a reverse engineering technique that helps in avoiding this issue. However, traditional software remodularization techniques primarily rely on structural dependencies and they generally neglect other important software features such as conceptual semantics and runtime behavior of a software system. This paper presents a Genetic Algorithm (GA)-based multi-dimensional remodularization framework that integrates structural, semantic, and runtime behavioral dependencies available in a software system. The framework evaluates seven dependency configurations arising from structural, semantic and behavioral dependency relations viz structural-only, semantic-only, behavioral-only, three pairwise combinations, and full integration. The evaluation of the remodularized solution is performed using three newly proposed quality metrics viz MQ+, SC, and CLA. Based on the carried out experimental evaluation, we demonstrate that full integration improves MQ+ by up to 25% over structural-only clustering while reducing cognitive load by nearly 40%.