An Improved Firefly Algorithm Based Cluster Analysis Technique

Abstract

Unsupervised machine learning approach like cluster analysis finds a large number of applications in different engineering domains. A variety of meta-heuristic algorithms have been proposed in the literature for clustering. Firefly is one of the most commonly used meta-heuristic algorithm as it has efficient capability of automatic subdivision of population and natural capability of dealing with multimodal optimization. But due to more dependency on local solution for movement, it generally leads to premature convergence. In this paper, an improved variant of firefly algorithm is proposed by introducing a new position updating equation for movement of firefly by using the idea of best solution for global search. A mutation operator is also incorporated in the basic firefly algorithm to enhance its convergence speed and exploration capability. The proposed firefly algorithm is simulated and compared with standard firefly algorithm on standard 13 benchmark functions. Moreover, the efficiency of the proposed firefly algorithm is also tested by adopting it as a clustering technique. The performance is tested on seven real-life datasets and also compared with various state-of-the-art meta-heuristic clustering techniques. The computation outcomes showed that the proposed algorithm is better in finding the optimal cluster center with minimum intra-cluster distance, along with fast convergence speed. Results are also verified quantitatively using Friedman, Wilcoxon, and post-hoc pairwise Nemenyi tests.