A particle swarm optimization approach for large-scale many-objective software architecture recovery

The software systems with a well-documented architecture are easy to understand and evolve. However, in most cases, either the documented architecture is unavailable or get eroded badly. In these cases, to understand and evolve the software systems, developers often need to recover the architectural compo-nents from the system implementation code. Recently, a variety of heuristic-based multi-objective opti-mization algorithms (MoOAs) for software architecture recovery (SAR) have been introduced. Most of the existing SAR approaches are designed by adopting the traditional MoOAs. However, the performance of such approaches degrades drastically with large-scale many-objective SAR (LSMaO-SAR). To address the challenges the MoOAs caused by the LSMaO-SAR, we introduce a large-scale many-objective particle swarm optimization (LSM-PSO) by customizing the framework of the PSO algorithm. For this, we adopt various strategies such as Balance Fitness Evaluation (BFE), Quality Indicator (QI) based fitness evalua-tion, Fuzzy-Pareto dominance (FPD), and Two-archive external storage, and incorporate into the PSO model. To test the effectiveness of the LSM-PSO, it is applied over five software projects and compared with the existing SAR approaches. The results show that the proposed LSM-PSO outperforms the existing optimization-based SAR approaches.(c) 2021 The Author. Published by Elsevier B.V. on behalf of King Saud University. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).