An Integrated Approach to High-Dimensional Data Clustering

Abstract

Applying the traditional clustering algorithms on high-dimensional datasets scales down in the efficiency and effectiveness of the output clusters. H-K Means is advancement over the problems caused in K-means algorithm such as randomness and apriority in the primary centers for K-means, still it could not clear away the problems as dimensional disaster which is due to the high-computational complexity and also the poor quality of clusters. Subspace and ensemble clustering algorithms enhance the execution of clustering high-dimensional dataset from distinctive angles in diverse degree, still in a solitary viewpoint. The proposed model conquers the limitations of traditional H-K means clustering algorithm and provides an algorithm that automatically improves the performance of output clusters, by merging the subspace clustering algorithm (ORCLUS) and ensemble clustering algorithm with the H-K Means algorithm that partitions and merge the clusters based on the number of dimensions. Proposed model is evaluated for various real datasets.