2009 IEEE International Conference on
Systems, Man, and Cybernetics |
![]() |
Abstract
A technique for accepting or rejecting the output of a Bayesian classifier is presented based on a metalevel Bayesian classifier using features of the probability distribution over the known classes calculated by the lower-level (object-level) classifier. The value of adding the metalevel classifier is calculated based on user utilities for the possible outcomes of the two-level process.