Method for enhancing performance of Bayesian network classifier using mutual information and recording medium for recording the method

摘要

The present invention is lean as a way to improve the Bayesian model averaging efficiency for multi-node order of beige anmang classifier that can improve the classification performance of the data, based on a number of variables set U with the training data D class variable (C Selecting feature variables that are deeply related to the; Generating a predetermined number of node orders { ₁ , ₂ ,..., _T } from the post probability distribution P (| D )) in consideration of only the selected feature variables; Calculating a coupling probability distribution ( P ( U | D )) for the node order using a Markovchain Monte Carlo method; Approximating the joint probability distribution for Bayesian classification; And applying the approximated bond probability distribution to a Bayesian classifier.;According to the present invention, it is possible to improve the execution speed of the Bayesian network classifier through Bayesian model averaging for the multi-node order without degrading the classification accuracy of the lean data.;Bayesian network classifier, feature selection, mutual information quantity, multi-node order, Bayesian model averaging, lean data