贝叶斯网络杂交学习算法及其在中医中的应用

Bayesian network approach to knowledge discovery in traditional Chinese medicine

针对贪婪贝叶斯模式搜索算法(GBPS)在搜索最优贝叶斯网络结构时易陷入局部最优的不足,提出了一种改进的GBPS算法.在GBPS算法的邻域生成过程中引入了有向边的变向操作,并通过仿真实验研究了样本数量和网络节点的连接边数对算法寻优能力、结果准确度和计算量的影响.将该改进算法用于从中医临床诊断数据中辨识症状与辨证要素间的复杂关系.结果表明,该改进算法的学习结果优于GBPS算法和贪婪贝叶斯有向无环图搜索算法(GBDS).所发现的症状-辨证要素间的相关关系与中医专家经验吻合较好,可用于从中医诊断数据中自动获取中医专家知识.

To overcome the local minimum of Bayesian network hybrid learning algorithm--greedy Bayesian pattern search algorithm （GBPS）, an improved algorithm was proposed by introducing the operation of reversing the directed edges in the generation of pattern search space. The influences of sample size and network node linkage number on the optimization ability, accuracy and computational cost were studied by using simulation experiments. Then the improved algorithm was applied to knowledge discovery from clinical data in traditional Chinese medicine （TCM）. The experimental results showed that the improved algorithm can yield more optimal and accurate Bayesian network structures than GBPS and another hybrid learning algorithm--greedy Bayesian DAG search algorithm （GBDS）. The independent and dependent relationships among symptoms and key elements for syndrome differentiation identified by the improved algorithm are very consistent with expert knowledge; and the algorithm can be used for acquiring knowledge for the construction of expert systems in TCM.