For the lack of effective method on multi-class’ probabilities modeling, a posterior probability estimating approach
based on error correcting output codes(ECOC) is presented. In this approach, probability estimation is translated into
solving overdetermined linear equations with the binary ECOC as a modeling framework. Moreover, when the modeling
framework is changed to ternary ECOC, the equivalence problem is overdetermined nonlinear equations. Therefore, an
iterative algorithm is proposed to solve this problem for obtaining probabilities. Compared with three classic approaches, the
proposed approach has better classification ability and probability distribution of the posterior probability in experiments.

pattern recognition;multi-class classification;probability modeling;error-correcting output codes