CLASSIFICATIONS OF CONTINUOUS-TIME MARKOV SOURCES AND AN APPLICATION TO ANALYSIS OF ELECTROENCEPHALOGRAMS (EEG)

A Bayesian approach for classification of Markov sources is developed and studied. Each of M sources is described by a continuous-time, disc rete-state Markov chain. All states and times of transitions between s tates can be observed perfectly but the transition rate matrices which establish the parameters of the sources are not known a priori. A Bay esian training algorithm using a fixed amount of memory digests the tr aining samples that consists of a member function from each chain. Thi s leads to an iterative and computationally simple classification and training algorithms. It is also shown the convergency of the parameter posterior density to the true value of the unknown parameters. Finall y, the algorithms established are applied for classification of Electr oencephalograms (EEG). Details of the initial data reduction, primitiv e feature selection which translate an EEG record into Markov states a nd the lengths of the sojourns in each state are explained. The experi mental results on the real data demonstrates that this approach can pr ovide substantial help to the clinicans and researchers who required E EG analysis.