基于共轭先验分布的贝叶斯网络分类模型

Bayesian network classifier based on conjugate prior distribution

针对贝叶斯网络后验概率需计算样本边际分布,计算代价大的问题,将共轭先验分布思想引入贝叶斯分类,提出了基于共轭先验分布的贝叶斯网络分类模型,针对非区间离散样本,提出一种自适应的样本离散方法,将小波包提取模拟电路故障特征离散化作为分类模型属性,仿真验证表明,模型分类效果较好,算法运行速度得以提高,也可应用于连续样本和多分类的情况,扩展了贝叶斯网络分类的应用范围。

In order to reducing calculate costs of Bayesian network, when calculating posterior probability of samples that need the marginal distribution, an approach of Bayesian network classifier based on conjugate prior distribution is proposed. An adaptive discretization method is also proposed to discrete non-interval samples. The fault feature of analog circuit extracted by wavelet packet is taken as a discrete property of Bayesian network classification model. The simulation result shows that, this classifier has high accuracy and efficiency of analog circuit fault diagnosis, and can be applied to continuous and multi-classification case, which extends the scope of application of Bayesian network classification.