Estimation of each q value for multi-q extension of Tsalli entropy based FCM

Abstract

By combining fuzzy c-means clustering, the Tsallis entropy maximization method, and deterministic annealing, we have developed the single-q clustering algorithm. Then, the algorithm has been extended to the multi-q clustering algorithm. In this method, the qs are assigned individually to each cluster. Each q value is determined so that the membership function fits the corresponding cluster distribution. This is done by introducing a new parameter. However, the accuracy of clustering depends on the setting of a value of the parameter. Accordingly, in this study we propose a new clustering method that does not require an additional parameter to determine the q values. Experiments are performed on randomly generated numerical data and "SDSS quasar spectra" dataset, and it is confirmed that the proposed method works correctly and improves the accuracy of clustering and is superior to the conventional multi-q method.