基于Markov blanket的无约束型K阶贝叶斯集成分类模型

Ensemble of unrestricted K-dependence Bayesian classifiers based on Markov blanket

为了提升K阶依赖贝叶斯分类(KDB)模型的条件依赖表达能力,本文以Markov blanket的特征提取思想为基本原则,降低特征属性间的条件独立性,根据贪婪搜索策略进行贝叶斯分类模型的结构学习。基于训练样本集构建宏观模型,基于测试样本构建微观模型,最终通过集成模型进行决策。针对UCI机器学习数据集进行交叉验证,实验结果分别从0-1损失、偏差和方差等角度证明了本文算法的合理性和有效性。

In order to promote the conditional dependence expression ability of K-dependence Bayesian Classifier,in this paper we relax the conditional independence between predictive features according to the basic idea of Markov blanket for feature extraction,and carry out structure learning based on greedy search strategy.We propose a macro model from the training set,and a micro model from each test sample.The decision is made by the ensemble of both models.We select data sets from UCI machine learning repository for cross validation.The experimental results demonstrate the rationality and effectiveness of the proposed algorithm in terms of 0-1 loss,bias and variance.