Performance evaluation of a novel hybrid clustering algorithm using birch and K-means

Clustering is a supreme descriptive task in data mining. Clustering is a way of grouping or combining data objects into disjoint clusters according to some criteria that you pick such that the data objects in the same clusters are similar and the data objects belonging to different clusters differ from each other. The clustering results directly depend upon the method of clustering algorithm that is applied. This research paper proposes a new hybrid method which combines the features of K-means clustering algorithm and BIRCH (a hierarchical clustering algorithm). The proposed algorithm firstly generates a tree using hierarchical clustering algorithm which gives a large number of clusters when applied to a dataset and then clustering has been performed using K-Means partitioning algorithm which reduces the number of cluster with more accuracy and less sum of square error. The proposed algorithm is applied on banking dataset which is then compared with K-means and K-Medoid clustering algorithm. The comparison is done on the basis of number of iterations and sum of square error (Intra cluster similarity), in which the new algorithm performs better as compare to K-Means and K-Medoid clustering algorithms.