Dynamic frequency based parallel k-bat algorithm for massive data clustering (DFBPKBA)

In the past one decade there has been significant increase in the growth of digital data. Therefore, good data mining techniques are important for the better decision making. Clustering is one of the key element in the field of data mining. K-means is a very popular algorithm present in the literature which is widely used for the clustering purpose. However k-means algorithm suffers from the problem of stucking into local optimum solution because of it’s dependency on the random initialization of initial cluster center. In this paper a novel variant of Bat algorithm based on dynamic frequency is introduced. Further the proposed variant is hybridized with K-means to present a new approach for clustering in distributed environment. Since evolutionary computation is very computation intensive, traditional sequential algorithms are not able to provide satisfactory results within the reasonable amount of time for the large scale data problems. To mitigate this problem the proposed variant is parallelized using the MapReduce model in the Hadoop framework. The experimental results show that the proposed algorithm has outperformed K-means, PSO and Bat algorithm on eighty percent of the benchmark datasets in terms of intra-cluster distance. Further DBPKBA has also achieved significant speedup for dealing with massive datasets with increase in the number of nodes.