Clustering on Big Data Using Hadoop MapReduce

Abstract

With the phenomenal increase in digital data, it is inefficient to run the traditional clustering algorithms on separate servers. To deal with this problem, researchers are migrating to distribute environment to implement the traditional clustering algorithms, more specifically K-means clustering. In traditional K Means Clustering, the problem of instability caused by the random initial centers exists. With random initial centres, if we execute the clustering algorithm (More specifically K-Means) on the same data set, more than once, we get different cluster results each time. Thus making the results unstable. Here, we proposed a modified K-Means clustering algorithm, which take the optimized initial centres based on data dimensional density. This approach deal with the random initial centers taken for algorithm execution and provides stable cluster results.