Performance analysis of firefly algorithm for data clustering

Extraction of relevant information from web is gaining immense significance. This depends upon the efficacy of methods employed to represent, organise and access the information. The most important technique that is used to organise the data is clustering. Clustering problem refers to partitioning unlabeled data objects into certain number of clusters with the objective of maximum homogeneity within cluster and heterogeneity between the clusters. The paper studies viability of firefly algorithm for clustering. It incorporates the flashing behaviour of fireflies to achieve the optimal solution. The performance of the proposed algorithm FClust is compared with particle swarm optimisation (PSO) and differential evolution (DE) algorithms with respect to varied statistical criteria using artificial and benchmark datasets. A detailed convergence behaviour of algorithms is studied using run length distribution. The experimental results prove that the proposed algorithm performs better in terms of speed and success rate as compared to PSO and DE.