Entropy-Type Classification Maximum Likelihood Method

M.-S. Yang and C.-Y. Lai (Taiwan)

Keywords

Classification maximum likelihood (CML), fuzzy clustering, entropytype CML (ECML )

Abstract

Mixtures of distributions are popularly used as probability models to analyze grouped data. Classification maximum likelihood (CML) is an important maximum likelihood approach to clustering for mixture models. Yang et al. had extended CML to fuzzy CML (FCML). Although the FCML presents better results than the CML, it is always affected by the parameter of fuzziness index. In this paper, we consider the fuzzy-type CML with an entropy regularization term and then create an entropy-type CML (ECML). The proposed ECML is a parameter-free algorithm. The comparisons of ECML with FCML and fuzzy clustering algorithms are made. These numerical and real data comparisons show that the proposed ECML algorithm presents better clustering results.

Important Links:



Go Back