A NEW SEQUENTIAL CLASSIFIER USING INFORMATION CRITERION WINDOW

We propose a model-based parametric unsupervised sequential classifier with an information criterion window (ICW), based on the assumption t hat the number of clusters changes smoothly from a stage to the follow ing stage, to determine the number of clusters at each stage and to pe rform classification on the observed mixture data. The proposed approa ch provides possible solutions for two problems of pattern recognition : adaptive classification and sequential clustering validation. Simula tion results demonstrate that the proposed algorithm provides good res ults for pure mixed data and successfully performs clustering for the mixed Gaussian data in the large-sample limit.