SCA2: Novel Efficient Swarm Clustering Algorithm

Clustering is a classical unsupervised learning task that aims to reveal data similarity patterns. Numerous algorithms have been proposed to address this task from different aspects. In the field of swarm intelligence and evolutionary algorithms, most existing algorithms strive to identify a set of cluster centers. However, it is difficult for centroid-based algorithms to process data with clusters of arbitrary shapes. Thus, a clustering algorithm named Swarm Clustering Algorithm (SCA) was proposed to cluster data from a novel aspect, which regards each point in the dataset as a particle, and particles fly towards denser areas to form clusters automatically. In this article, a novel efficient swarm clustering algorithm named SCA2 is proposed, which extends SCA in terms of three aspects: (1) the radial basis function network is adopted as the surrogate model to reduce the time complexity; (2) there are $k$ leaders for each particle, and the particle may follow one of them to decrease misleading; and (3) a simplified strategy is used to update the position of each particle. The performance of SCA2 on different types of synthetic and real-world datasets was compared with the performance of four classical algorithms, SCA as well as a PSO-based clustering algorithm. The experimental results demonstrate that SCA2 is more competitive.