离散属性的朴素贝叶斯分类算法的优化  

Optimization of Naive Bayesian Classification Algorithm for Discrete Attributes

作　　者：李福祥 王建敏 梁建创 王雪 LI Fu-xiang;WANG Jian-min;LIANG Jian-chuang;WANG Xue(College of Science,Harbin University of Science and Technology,Harbin 150080,China;Technology R&D Department,Hitrendtech,Shanghai 200000,China)

机构地区：[1]哈尔滨理工大学理学院,哈尔滨150080 [2]钜泉光电科技(上海)股份有限公司技术研发部,上海200000

出　　处：《小型微型计算机系统》2022年第5期897-901,共5页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(11871181)资助;黑龙江省自然科学基金项目(A2018008)资助。

摘　　要：朴素贝叶斯算法是一种经典的分类算法,广泛应用于很多领域.朴素贝叶斯分类算法引入了属性条件独立性假设,但这个假设在现实应用中往往不能满足,从而就会影响算法的分类性能.针对这一问题,本文对该算法进行了改进,对离散属性进行数值标记,之后用正交矩阵对连续属性和数值标记后的离散属性做正交变换,增强属性之间的相互独立性,去除了属性之间的线性关系,贴近了朴素贝叶斯分类算法的属性条件独立性假设,从而提高了分类准确率.最后基于改进的算法进行实验分析,实验结果表明,与标准朴素贝叶斯分类算法、贝叶斯网相比,改进的算法的分类性能有较大的提高.Naive Bayes algorithm is a classical classification algorithm,which is widely used in many fields.Naive Bayes classification algorithm introduces the assumption of attribute condition independence,but this assumption is often not satisfied in practical application,which will affect the classification performance of the algorithm.In order to solve this problem,this paper improves the algorithm by marking discrete attributes with numerical marker,and then uses orthogonal matrix to transform continuous attributes and discrete attributes after numerical labeling with orthogonal matrix to enhance the mutual independence of attributes,remove the linear relationship between attributes,and approach the assumption of attribute condition independence of naive Bayes classification algorithm,so as to improve the classification Accuracy.The experimental results show that compared with the standard naive Bayes classification algorithm and Bayesian network,the improved algorithm has better classification performance.

关 键 词：朴素贝叶斯分类 数值标记 正交矩阵 属性独立 十折交叉验证 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]