数据缺失下学习贝叶斯网络的E-GSA算法

摘要

针对数据缺失条件下构建贝叶斯网络难度大的问题,研究了贝叶斯结构学习算法,提出了将条件独立性检验和评分-搜索相结合的算法.采用改进的混合算法对训练数据初始化,建立相应的初始网络,对已经拟合了训练数据信息的初始网络用遗传模拟退火算法进行训练以找到最佳的网络结构.给出了算法实施的具体步骤且通过实验验证了算法性能,并将实验结果与其他典型的算法进行比较,表明了算法具有更优的学习效果.%For solving the difficulty of building Bayesian network with missing data,structural learning algorithm of Bayesian network is studied,an algorithm combines conditional independence test and score-search is presented.First,the initial network is built by the initialization of the training data using improved hybrid algorithm.Then make use of genetic-simulated annealing algorithm to train the initial network which has combined the training data in order to find the best network.Detailed operation steps are given out and the algorithm is also compared to other well-known algorithms.Experimental results indicate that this algorithm makes a more effective study performance than several other algorithms.