| 简介 |
PURPOSE : A method for efficient Bayesian model averaging of Bayesian network classifiers over multiple node-orders via feature selection based on normalized mutual information is provided to increase a processing time without decreasing the classification accuracy for the rare data of Bayesian network classifier through Bayesian model averaging for multi-node sequence.CONSTITUTION : Based on plural variable sets and learning data, a feature variable deeply related to a class variable is selected. The node sequence is generated from post probability distribution by considering only selected feature variable. By using a Markvo-chain Monte-Carlo method, a combined probability distribution for node sequence is calculated. A combination probability distribution is approximated for the Bayesian classification, and the approximated combination probability distribution is applied to the Bayesian classifier.© KIPO 2009 |
客服热线：
手机版
二维码
