产生式与判别式分类方法比较研究  

A Comparative Study of Generative and Discriminative Classification Methods

作　　者：孔环环[1] 石洪波[1] 

机构地区：[1]山西财经大学信息管理学院,山西太原030006

出　　处：《科技情报开发与经济》2010年第8期78-81,共4页Sci-Tech Information Development & Economy

基　　金：国家自然基金项目(60503017);山西省自然基金项目(20051046)

摘　　要：分类是数据挖掘领域研究的热点,产生式与判别式是数据挖掘中两种不同的分类模型。产生式模型具有通用性、灵活性及清晰的分层结构,学习得到的模型很容易满足模型解释要求;判别式模型没有明显的对系统中变量的基本分布建模的企图,仅仅对输入到输出之间映射的最优化感兴趣,可以提供更好的分类性能。从准确率、建模时间及渐进误差等方面对产生式与判别式分类方法进行了分析与比较,为研究人员在分类模型的选择上提供了参考。Classification is a hot research in the field of data mining,and the generative model and the discriminative model are two different types of classification models in data mining.The generative model has the versatility,flexibility and clear hierarchical structure,learning the obtained model can easily meet the need of model interpretation;the discriminative model,which has no obvious attempts to model the fundamental distribution of variables of the system,is only interested in the optimization of the mapping between input and output,which can provide better classification performance.This paper makes analysis and comparison of the two classification methods from the accuracy,modeling time,and incremental error,etc.for providing reference for the researchers in the choice of the classification model.

关 键 词：分类 判别式模型 产生式模型 支持向量机 朴素贝叶斯 

分 类 号：TP274[自动化与计算机技术—检测技术与自动化装置]