基于贝叶斯网络的海量数据多维分类学习方法研究

摘要

为了提高多维分类的执行效率，同时保持高的预测准确性，提出了一种基于贝叶斯网络的多维分类学习方法。将多维分类问题描述为条件概率分布问题。根据类别向量之间的依赖关系建立了条件树贝叶斯网络模型。最后，根据训练数据集对条件树贝叶斯网络模型的结构和参数进行学习，并提出了一种多维分类预测算法。大量的真实数据集实验表明，提出的方法与当前最好的多维分类算法 MMOC 相比，在保持高准确性的同时将模型的训练时间降低了两个数量级。因此，提出的方法更适用于海量数据的多维分类应用中。%In order to improve the execution efficiency of multi-dimensional classification while preserving high prediction ac-curacy,this paper proposed a Bayesian net based multi-dimensional classification learning algorithm.Firstly,it described the problem of multi-dimensional classification as the problem of conditional probability distribution.Secondly,it built a condi-tional tree Bayesian net model according to the dependence of class vector.Finally,it learnt the structure and parameters of the conditional tree model based on the training data set,and proposed a multi-dimensional classification prediction algorithm. Massive experiments on real dataset show that,compared with the state-of-the-art multi-dimensional classification algorithm MMOC,the proposed algorithm improves the execution efficiency of multi-dimensional classification while preserving high pre-diction accuracy.So,the proposed algorithm is more suitable in multi-dimensional classification for massive data.