IAFCO: an intelligent agent-based framework for combinatorial optimization

Solving combinatorial optimization problems (COPs) poses a significant challenge in various application domains. The NP-hardness of many COPs necessitates the integration of meta-heuristics to effectively tackle these problems by leveraging the strengths of each meta-heuristic. However, hybrid meta-heuristics lack a mechanism to determine when to activate specific techniques during the search process. To address these limitations and enhance the overall generality, this study proposes a novel agent-based hyper-heuristic framework, referred to as IAFCO, for solving COPs. The proposed framework imbues the search strategy with greater intelligence and information, resulting in improved performance. The proposed framework comprises 4 coalitions that leverage different search perspectives to enhance the search process: a global coalition, a local coalition, a learning coalition, and a perturbation coalition. Additionally, a control agent is introduced to coordinate the coalitions. Using a reinforcement learning mechanism, the control agent autonomously selects the most suitable coalition of agents based on the search state. The set of states employed in this study is problem-independent and encompasses all possible states that may arise during the search process. Furthermore, some agents adopt chaotic numbers instead of random numbers to maintain diversity. To showcase the effectiveness of the proposed framework, its performance is evaluated on the software modularization problem (SMP) -also known as software clustering- and community detection problem (CDP) in complex networks as 2 case studies. The experiments on SMP's software systems, including 10 applications and 10 folders of Mozilla Firefox, and comparison with 21 modularization algorithms and experiments on CDP's networks, including 9 real-world networks, and comparison with 22 state-of-the-art community detection algorithms, indicate that the proposed framework is comparable to other existing methods. The source codes and datasets of this paper are available at https://github.com/mahjoubeh-t/IAFCO.