PARALLEL META-HEURISTIC ALGORITHM FOR CLUSTERING

Abstract

Data mining involves the extraction of hidden patterns from raw data. Knowledge mining from raw data is data mining. Cluster Analysis is an important part of Data Mining activities. Analyzing the similarity in data helps to find many useful patterns of data to find relevant results. Clustering finds application in many areas such as pattern analysis, decision-making, grouping and machine-learning examples, including document retrieval, data mining, pattern classification and image segmentation. Today, the amount of data available has increased manifold. The analysis of this large amount of data is a computationally intensive task and requires a lot of execution time. Hadoop MapReduce is a new framework that allows parallelization, fault tolerance, node balancing and data distribution in library. It consists of user-defined map and reduce tasks. Use of MapReduce for clustering has seen a rise recently so that large datasets can be mined easily. Likewise, in this thesis we present a Parallel Meta-Heuristic Algorithm for clustering large amount of data. A K-Bat algorithm has been developed that uses advantages to two traditional algorithms and gives comparably good results. The proposed method uses the dynamic exploration and exploitation capability of the bat algorithm. It removes its defect of inappropriate timing of exploitation activity. It uses K-means structure to give initial population to the proposed algorithm. The parallel structure of this algorithm is then proposed. After testing it on benchmark datasets, it shows extremely efficient performance. It achieves best fitness results as well as best execution time.