基于互信息的贝叶斯网络结构学习

Bayesian network structure learning based on mutual information

利用贝叶斯网络对数据建立模型是目前人们研究热点。结构学习是贝叶斯网络的难点之一。为了避免启发式算法陷入局部最优和出现退化现象,提出了修正非法结构的人工免疫优化算法,通过适应度评分机制提取全局最优疫苗并对种群个体进行疫苗注射,采用评分机制修正有环图和双向边的非法贝叶斯网络结构。得到的网络结构具有较高的适应度。通过对经典网络进行结构仿真,验证了算法的效率和准确性。

Data modeling based on Bayesian network has received tremendous attention at present. Structure learning is one of the main pain points in Bayesian network. In order to avoid local optimal and degradation, an artificial immune algorithm is proposed to improve the illegal structure. First, the global optimum vaccine is extracted by a fitness function to modify the illegal structures. Then the proposed algorithm is compared with other learning algorithms for typical Bayesian networks. Experimental results show that the proposed algorithm is effective and accurate.