A Novel History-driven Artificial Bee Colony Algorithm for Data Clustering

Clustering is a branch of unsupervised learning in which data are divided into groups with similar members called cluster. Since clustering is considered as a NP-hard problem, many swarm intelligence-based algorithms have been applied in recent years to solve this problem effectively. In this paper, a novel variant of Artificial Bee Colony (ABC) algorithm called History-driven Artificial Bee Colony (Hd-ABC) is proposed to improve the ABC's performance by applying a memory mechanism. The Hd-ABC utilizes a binary space partitioning (BSP) tree to memorize useful information of the evaluated solutions. Applying this memory mechanism, the fitness landscape can be approximated before actual fitness evaluation. Fitness evaluation is a costly and time consuming process in clustering problem, but utilizing the memory mechanism has decreased the number of fitness evaluations significantly and accelerated the optimization process by estimating the fitness value of solutions instead of calculating actual fitness values. Moreover, inspired by guided anisotropic search (GAS) strategy, a new local search mechanism is introduced to improve the exploitation capability as well as the convergence speed of the ABC in the onlooker bee phase. In the proposed algorithm, the GAS strategy incorporates the BSP tree to mutate solutions in the direction of the nearest optimum instead of the random walk of the ABC. Also, in order to improve the global search strategy of the ABC in the scout bee phase, a new mechanism is proposed which produces fitter starter solutions by finding and modifying the worst dimension in each solution. The proposed algorithm, for data clustering has been applied on nine UCI datasets and two artificial datasets. Both the experimental and statistical results show that the proposed algorithm outperforms the original ABC, its variants and the other state-of-art clustering algorithms; and the simulations indicate very promising results in terms of solution quality.