Abstract: The concept of reasoning information is presented, which is used as the measure of discretization of continuous variables in Bayesian Network(BN). Genetic algorithm is used to search the best solution. Encoding method, crossover operator and mutation operator is proposed. Reasoning information is used as the function of individual fitness. Experiment proves that the Bayesian Network learning from data based on this discretization method can get more reasoning information.

Key words: parameter learning, reasoning information, discretization method, genetic algorithm

摘要： 提出推理信息量的概念，将其作为贝叶斯网络连续变量离散化评价标准。在连续变量离散化的过程中，采用遗传算法寻求最优解，设计个体编码方式、交叉算子和变异算子，将推理信息量作为衡量个体适应度的标准。实例分析证明，通过该方法对变量进行离散化后学习得到的贝叶斯网络在推理时能得到更大的推理信息量。

关键词: 参数学习, 推理信息量, 离散化方法, 遗传算法

CLC Number: 

• TP393