An Efficient Automatic Bat Clustering Algorithm for Multi Objective Optimization

Abstract

In data mining the clustering techniques is used for grouping a set of physical or abstract objects into similar objects. In this process, k-means algorithm is a major role to group the similar objects. The major issue of this algorithm is the user gives the number of clusters in priori as k value where as the final clustering results is ineffective. To avoid such a problem a new Multi Objective (MO) method Bat Modified Clustering Multi-Objective Optimization (BATMClustMOO) is proposed. This algorithm is a combination of Archived Multi-Objective Simulated Annealing (AMOSA) and Bat algorithm (BA) is suggested which can partition the data into a suitable number of clusters k and then find the best cluster centroid automatically. The AMOSA acts as the local search and BA acts as the global search to fix the number of clusters and cluster centroid. Each cluster is splitted into many small hyper spherical sub clusters and the centroid of all small sub-clusters is fixed into a string that comprises the entire clustering. In order to verify the performance of the proposed algorithm the different benchmark datasets are taken from UCI repository. The experimental results show the proposed method is better than the existing methods.