The Algorithm of Deviation Measure for Cluster Models Based on the FOCUS Framework and BIRCH

The work of how to measure the difference, or deviation, between two models is an important studied problem. Venkatesh Ganti has developed the FOCUS framework for computing an interpretable, qualifiable deviation measure between two datasets in terms of the models they induce. In this paper, we propose a novel method to measure the deviation for cluster models making use of CF-tree based on the FOCUS framework and BIRCH algorithm. First, for the different comparable datasets, we build the CF-tree as the GCR of different structural components. Second, we use geometrical distance that we defined to quantify the difference between datasets. An experiment of the algorithm on the 2-dimensional synthetic datasets is presented, demonstrating the applicability of the proposed algorithm.