Clonal Selection based Parallel Fuzzy Clustering using Map-reduce

摘要

In recent decade, clustering is an important task in data mining in which data is grouped into different clusters. The clustering of data can be done using either crisp or fuzzy clustering algorithms. It has been seen that fuzzy based approaches are more reliable than crisp clustering techniques in terms of accuracy. But the random centroid initialization in fuzzy clustering leads solution to converge at local optima. Also the amount of data is endlessly increasing nowadays; hence to cluster the large volume of data is of major concern. In this paper, we consider both the above issues and proposed a parallel fuzzy clustering algorithm using clonal selection principle. The problem of convergence at local optima is resolved by using clonal selection principle and clustering of large datasets can be done using distributed map reduce framework. A scalable library mahout that runs on top of Hadoop is used for parallel fuzzy clustering. The experimental analysis is done on multi node Hadoop cluster and validated using different datasets.