Research Area: Big Data
With advancement of the technology, data size is increasing rapidly. For making intelligent decisions based on data, efficacious analytic methods are required. Data clustering, a prominent analytic method of data mining, is being efficiently employed in data analytics. To analyze massive data sets, the improvement in the traditional methods is the urge of todays scenario. In this paper, an efficient clustering method, MapReduce based enhanced grey wolf optimizer (MR-EGWO), is presented for clustering large-scale data sets. The proposed method introduced a novel variant of grey wolf optimizer, Enhanced grey wolf optimizer (EGWO), where the hunting strategy of grey wolf is hybridized with binomial crossover and lévy flight steps are inducted to enhance the searching capability for pray. Further, the proposed variant is used for optimizing the clustering process. The clustering efficiency of the EGWO is tested on seven UCI benchmark datasets and compared with the five existing clustering techniques namely K-Means, particle swarm optimization (PSO), gravitational search algorithm (GSA), bat algorithm (BA) and grey wolf optimizer (GWO). The convergence behavior and consistency of the EGWO has been validated through the convergence graph and boxplots. Further, the proposed EGWO is parallelized on the MapReduce model in the Hadoop framework and named MR-EGWO to handle the large-scale datasets. Moreover, the clustering quality of the MR-EGWO is also validated in terms of F-measure and compared with four MapReduce based state-of-the-art namely; parallel K-Means, parallel K-PSO, MapReduce based artificial bee colony optimization (MR-ABC), dynamic frequency based parallel k-bat algorithm (DFBPKBA). Experimental results affirm that the proposed technique is promising and powerful alternative for the efficient and large-scale data clustering.
Keywords:
Author(s) Name: Ashish Kumar Tripathi,Manju Bala and Kapil Sharma
Journal name: Big Data Research
Conferrence name:
Publisher name: ELSEVIER
DOI: 10.1016/j.bdr.2018.05.002
Volume Information: Volume 14, December 2018, Pages 93-100
