A learning automata‐based clustering algorithm using ant swarm intelligence. Issue 6 (24th July 2018)

• Record Type: Journal Article
• Authors: Anari, Babak; Akbari Torkestani, Javad; Rahmani, Amir Masoud
• Is Part Of: Expert systems. Volume 35:Issue 6(2018)
• Page Start: n/a
• Page End: n/a
• Publication Date: 2018-07-24
• Subjects: ant‐based clustering -- data clustering -- learning automata -- swarm intelligence; Expert systems (Computer science); 006.33
• DOI: 10.1111/exsy.12310
• Languages: English
• ISSNs: 0266-4720

Abstract

Abstract: Ant‐based clustering algorithms are inspired by the behaviour of real ants. In most ant‐based clustering algorithms, each ant lacks global visibility and only uses local search to find the best place on a grid to drop its data item. This paper presents a new clustering method called learning automata‐based clustering algorithm using learning automata (LA) and ant swarm intelligence. In this paper, the problem of finding the best location of grid for dropping a data item is solved using LA. We introduced a new drop operation using the global search capability of LA that can increase the quality of clustering. In the proposed method, each ant is equipped with an LA and a two‐dimensional grid is partitioned into a number of clusters. To drop a data item, the LA of each ant will be responsible for finding one of the best clusters on a grid unlike other methods that use the random search of a grid. To evaluate the results, some clustering evaluation criteria, such as the number of obtained clusters, f‐measure, Rand index, intracluster variance, Dunn index, error rate, and time cost were employed. Additionally, the proposed algorithms were compared with k‐means, Lumer and Faieta, ant clustering algorithm, modified version of the ant‐clustering algorithm, Chaotic Ant clustering (CAC), chaotic ant clustering algorithm, adaptive ant‐based clustering algorithm, adaptive artificial ant clustering algorithm, and ant clustering algorithm with information theoretic learning algorithms on both real and synthetic datasets. Experimental results show that the proposed method can increase the efficiency of the real dataset by 11.13% and that of the synthetic dataset by 16.11%. It also reduces the number of neighbouring function calls by 16% for the real dataset and by 28% for the synthetic dataset.