多贝叶斯网络分类器集成模型研究

Research on Integrated Model of Several Bayesian Network Classifiers

提出了一种多贝叶斯网络集成的分类和预测方法。把专家知识作为"疫苗",利用免疫遗传算法和约束信息熵适应度函数相结合的方法进行贝叶斯网络结构的学习,得到多个反映同一样本数据集的、网络结构复杂度折衷的、满意的贝叶斯网络结构。然后,给出了多贝叶斯网络分类器集成模型,把学习得到的贝叶斯网络进行集成,代表"专家"对未知类别的不完全数据进行群决策的分类和预测,提升贝叶斯网络分类器的泛化能力。最后,结合贝叶斯推理工具GeNIe软件,通过实例说明该方法的合理性和有效性。

A classification approach based on several Bayesian network classifiers integrated is presented. Bayesian network structure learning is implemented by combining immune genetic algorithm with information entropy restrained on adaptive function looking on expert knowledge as ＂bacterin＂, and several Bayesian network structures satisfying the same sample data collection are gained. Then, it shows integrated model of several Bayesian network classifiers. Each Bayesian network structure is integrated together, and represents expert to classify and forecast class of the incomplete data in order to improve generalizing performance of the whole Bayesian network Classifier. Finally, integrating with software tool GeNle, it gives an example to illustrate the rationality and validity of the approach.