Fast isodata clustering algorithms

摘要

The computational requirements of any clustering method are identified as the major bottleneck in the effective exploratory data analysis task. Partial sum and nearest neighbouring distance methods are proposed to speed up the K-MEANS clustering algorithm with Euclidean distance norm. Use of the Mahalanobis distance norm instead of the Euclidean distance norm is identified for the effective exploration of the real structure of the data. Lower triangular matrix and Expanded distance function approaches are proposed to speed up the Mahalanobis distance based K-MEANS algorithm. Further, use of the Dynamic clustering method for optimal global partition of the data is identified and use of the Lower triangular matrix approach to speed up Dynamic clustering is proposed. Other methods to speed up the Dynamic clustering process are identified. Use of the image auto-correlation property is identified for further speeding up of the proposed methods in the clustering of image data.