Abstract.A stochastic model for the description of the classification problem is presented. Statistically motivated supervised and unsupervised training schemes for classifiers are considered; the resulting classifiers turn out to be asymptotically optimal. The rates of convergence of probability of successful classification to optimality are studied in a computer experiment. The supervised training scheme entails a sequence of classifiers whose quality converges faster to optimality than that in the unsupervised case.

Received: December 20, 2009

AMS Subject Classification: 91E40, 62F12, 62G20

Key Words and Phrases: Bayesian classifier, EM algorithm, consistent estimator

Source: International Journal of Pure and Applied Mathematics
ISSN: 1311-8080
Year: 2010
Volume: 59
Issue: 2