基于马尔可夫毯的贝叶斯网络结构学习算法

贝叶斯网络图结构的自动学习是机器学习中的一个挑战，针对传统算法学习效率低、难于去除冗余边及确定结构中边的方向等问题，提出了一种基于马尔可夫毯的贝叶斯网络结构学习算法.该算法改进了经典的马尔可夫毯学习算法，使之减少条件独立检验次数，并在后续确定有向结构方面更适应贝叶斯网络结构学习，同时给出了两种有向边方向确定的一般性解决方案，有效提高了学习算法的学习效率.最后建立了基于贝叶斯网络的互联云QoE评价模型，并进行了仿真实验，结果表明改进后的学习算法在预测准确率、学习效率上均优于传统算法.

Structure Learning Algorithm of Bayesian Networks Based on Markov Blanket

The automatic learning of Bayesian network graph structure is a challenge in machine learning. Aiming at the problems of low learning efficiency of traditional algorithm， difficulty in removing redundant edges and determining the direction of the edges in the structure， a Bayesian network structure learning algorithm based on Markov blanket was proposed. The proposed algorithm improves the classical Markov blanket learning algorithm， reduces the number of conditional independent inspections， and is more suitable for Bayesian network structure learning in the subsequent determination of directed structures. At the same time， a general solution for determining the direction of two directed edges was given， which effectively improves the learning efficiency of the learning algorithm. Finally， the Bayesian network-based interconnected cloud QoE evaluation model was established， and the simulation experiment was carried out. The results showed that the improved learning algorithm is superior to the traditional algorithm in prediction accuracy and learning efficiency.