MICSA: a multi-strategy integrated chameleon swarm algorithm for community detection with a new objective function

• Published in: Neural computing & applications Vol. 37; no. 22; pp. 17887 - 17912
• Main Authors: Xie, Wentao, Kang, Yan, Huang, Xin, Yang, Mingjian, Yuan, Hu
• Format: Journal Article
• Language: English
• Published: London Springer London 01.08.2025; Springer Nature B.V
• Subjects: Algorithms; Artificial Intelligence; Clustering; Computational Biology/Bioinformatics; Computational Science and Engineering; Computer Science; Connectivity; Data Mining and Knowledge Discovery; Deep learning; Graph theory; Heuristic methods; Image Processing and Computer Vision; Modularity; Probability and Statistics in Computer Science; Regularization methods; Review; Search methods; Community detection; Swarm intelligence algorithm; Multi-objective optimization; Complex networks
• ISSN: 0941-0643; 1433-3058
• DOI: 10.1007/s00521-025-11266-7

Community detection (CD) unveiling the latent structural organization behind the complex networks holds substantial value in various applications. Although metaheuristic algorithms have been introduced to effectively address the CD, most methods are prone to prematurely converge to the local optimum. Furthermore, the common objective function such as modularity may fail to evaluate the diverse characteristic of communities. Therefore, we propose a multi-strategy integrated chameleon swarm algorithm (MICSA) for CD to effectively avoid local optima and provide fast convergence speed. Furthermore, a novel multi-objective function is presented by introducing the metric in directed graph of software module clustering to CD for further evaluating the various characteristic of communities. Additionally, various strategies are combined to generate high-quality initial population. An adaptive multi-granularity searching strategy implements different mutation operators according to converge. And elites prey searching strategy is proposed to utilize the community structure among elite individuals. These strategies are integrated with prey searching stage to escape from local optimum and enhance the exploitation capability. The vector similarity strategy utilized graph embedding is combined with prey capture stage to differentiate complex relationships between nodes. Extensive experiments in both types of networks demonstrate the superiority of MICSA compared with fifteen classical and state-of-the-art algorithms.