一种基于属性分割的产生式/判别式混合分类器

Generative/discriminative hybrid classifier based on attributes partitioning

为了利用产生式和判别式方法各自的优势,研究了基于属性分割的产生式/判别式混合分类模型框架,提出了一种基于属性分割的产生式/判别式混合分类器学习算法GDGA。其利用遗传算法,将属性集X划分为两个子集XG和XD,并相应地将训练集D垂直分割为两个子集DG和DD,在两个训练子集上分别学习产生式分类器和判别式分类器;最后将两个分类器合并形成一个混合分类器。实验结果表明,在大多数数据集上,混合分类器的分类正确率优于其成员分类器。在训练数据不足或数据属性分布不清楚的情况下,该混合分类器具有特别的优势。

In order to exploit the best of generative and discriminative approaches,on the basis of investigating a framework of generative/discriminative hybrid model based on attributes partitioning,this paper proposed a learning algorithm of generative/discriminative hybrid classifier based on attributes partitioning,GDGA.This algorithm divided the attributes X into two subsets,XG and XD,by applying genetic algorithms,and vertically partitioned the training set into two subsets DG and DD accordingly.Then it trained a generative classifier and a discriminative classifier on DG and DD respectively.In the final,it constructed a generative/discriminative hybrid classifier by combining the generative classifier and the discriminative classifier.Experimental results show that the generative/discriminative hybrid classifier performs better than its generative component and discriminative component on most data sets.This hybrid classifier has particular advantage in the case of unclear attribute distribution or not enough training data.