Fusion of K-Means Algorithm with Dunn's Index for Improved Clustering

Clustering is basically used for the determination of undetected patterns and the grouping of primary data which has to be processed. There are disjointed clusters and the partitions of data are done such that data of similar type are under the same cluster. Real time applications use K-means algorithm as K- means works out as an efficient method for the production of clusters. In this research we are focused on developing an algorithm keeping k-means as base algorithm but with the dynamic allotment of required clusters furthermore focusing on parallel approach, instead of executing the one by one. The greatly used K-Means increases problems because of declaration of cluster size in beginning. The computations between the data sets using the K-Means algorithm are performed step by step such that second starts only after first is completed. This work provides us a new technique for declaring dynamic clusters for the K-means method and further the Parallel Libraries are used for parallel computations of data. The computations are performed on 2-D primary data using the K-Means approach as we as the designed Modified K-Means. The results therefore validates that the Modified method is further more effective and efficient than the K-Means either in terms of production of clusters as Modified algorithm itself decides the required clusters claimed for parallel computation which improves the processing and decreases time. For the different datasets in database the results are computed using K-means and the modified method which show the %expedite of 20 to 50 percent when computations are done for same cluster size.