Abstract
Multi-objective software module clustering problem aims to automatically produce clustering solutions that optimize multiple conflicting clustering criteria simultaneously. Multi-objective evolutionary algorithms have been a most appropriate alternate for solving M-SMCPs. Recently, it has been observed that the performance of MOEAs based on Pareto dominance selection technique degrades with multi-objective optimization problem having more than three objective functions. To alleviate this issue for M-SMCPs containing more than three objective functions, we propose a two-archive based artificial bee colony algorithm. For this contribution, a two-archive concept has been incorporated in the TA-ABC algorithm. Additionally, an improved indicator-based selection method is used instead of Pareto dominance selection technique. To validate the performance of TA-ABC, an empirical study is conducted with two well-known M-SMCPs, i.e. equal-size cluster approach and maximizing cluster approach, each containing five objective functions. The clustering result produced by TA-ABC is compared with existing genetic based two-archive algorithm and non-dominated sorting genetic algorithm II over seven un-weighted and 10 weighted practical problems. The comparison results show that the proposed TA-ABC outperforms significantly TAA and NSGA-II in terms of modularization quality, coupling, cohesion, Pareto optimality, inverted generational distance, hypervolume, and spread performance metrics.