Neural network data clustering on the basis of scale invariant entropy

A new method for data clustering is proposed. The method uses generalized scale invariant concept of distance measure and data entropy. The analysis of analogy between known Euclidean metric and the proposed measure allows constructing an effective clustering algorithm. The developed technique enables grouping of heterogeneous data regardless of the measuring scale chosen and can be used in different applications. The demonstration examples of clustering Iris flower data and Wine recognition data are considered. New algorithm shows low error rate for the examined data sets surpassing traditional algorithms derived from Euclidean metrics, whereas simultaneously preserving the scale invariant property.