基于类约束的贝叶斯网络分类器学习

Learning Bayesian Network Classifiers Restricted by Class Variable

分类能力是人类经过学习得到的重要而基本的能力 ,也是机器学习、模式识别和数据采掘研究的核心问题 .在0 - 1损失率下 ,证明了基于类约束的贝叶斯网络分类器是最优分类器 .建立该分类器的核心问题是基于类约束属性贝叶斯网络结构学习 ,给出了学习属性贝叶斯网络结构的方法 ,在学习过程中使用了根据弧方向因果语义确定边方向的方法 ,并和碰撞识别定向相结合 ,在边定向之后进行冗余弧检验 ,解决了目前冗余边检验在定向之前所导致的问题 ,显著提高了结构学习效率和准确性 .并使用模拟数据进行了分类实验和分析 .

The classification is an important and basic ability for human obtained by learning. It has been considered as a key research area in machine learning, pattern recognition and data mining. It is proved that a Bayesian network classifier restricted by class variable is optimal under zero-one loss rate. The most important problem of setting up the classifier is to learning the structure of attributes Bayesian network restricted by class variable. In this paper, the method of learning the structure of attributes Bayesian network is developed. In learning the method of orienting edges based on the causal semanitics of an arc's direction is used. The method is combined with that of orienting edges based on collider identification to make superfluous arcs disposal after orienting edges. The problems brought by checking superfluous edges before orienting edges are avoided. The efficiency and veracity of learning Bayesian network structure is markedly improved. A contrast experiment is conducted by simulation and the results are analyzed.