Improved MapReduce k-Means Clustering Algorithm with Combiner

Data Mining and High Performance Computing are two broad fields in Computer Science. The k-Means Clustering is a very simple and popular data mining algorithm that has its application spread over a very broad spectrum. MapReduce is a programming style that is used for handling high volume data over a distributed computing environment. This paper proposes an improved and efficient method to implement the k-Means Clustering Technique using the MapReduce paradigm. The main idea is to introduce a combiner in the mapper function to decrease the amount of data to be written by the mapper and the amount of data to be read by the reducer which has considerably reduced the redundant Map-Reduce calls that have resulted in a significant reduction in the time required for clustering as it has decreased the read/write operations to a large extent. The implementation of Improved MapReduce k-Means Clustering has been clearly discussed and its effectiveness is compared to the regular implementation in an experimental analysis. The results consolidate this research by concluding that the Improved MapReduce Implementation of k-Means Clustering Algorithm out performs the regular implementation by over 300 seconds.