目前很多离散化方法仅考虑单个变量,不能得到最优的离散化方案.文中提出一种多属性关系的数据离散化方法.凭借概率的模型选择和最小描述长度原理,获得多变量离散化衡量标准,基于该标准提出一种有效的启发式算法来寻找最好的离散化方案.对UCI数据集进行分类预测,实验结果表明该方法提高Nave贝叶斯分类器的学习精度. At present, many discretization methods only consider a single variable and can not get the optimal discretization scheme. In this paper, a method of data discretization with multiple attribute relations is proposed. By virtue of the model selection of probability and the principle of minimum description length, a multivariate discretization criterion , An effective heuristic algorithm is proposed to find the best discretization scheme based on the standard.Classification and prediction of UCI dataset show that the proposed method can improve the learning accuracy of Nave Bayesian classifier.