摘要
提出了一种新的基于边缘分类能力排序准则,用于基于排序聚集(ordered aggregation,OA)的分类器选择算法。为了表征分类器的分类能力,使用随机参考分类器对原分类器进行模拟,从而获得分类能力的概率模型。为了提高分类器集成性能,将提出的基于边缘分类能力的排序准则与动态集成选择算法相结合,首先将特征空间划分成不同能力的区域,然后在每个划分内构造最优的分类器集成,最后使用动态集成选择算法对未知样本进行分类。在UCI数据集上进行的实验表明,对比现有的排序准则,边缘分类能力的排序准则效果更好,进一步实验表明,基于边缘分类能力的动态集成选择算法较现有分类器集成算法具有分类正确率更高、集成规模更小、分类时间更短的优势。
This paper proposed a new ordering criterion which could be used by classifiers selection algorithm based on or- dered aggregation. For calculating the competence of the classifier, it used a randomized reference classifier for modeling the classifier to get the probabilistic model of classifier competence. By combining with ordering criterion based on classifier com- petence of margin, the paper proposed a novel dynamic ensemble selection algorithm （CCM-DES） for improving the perform- ance of the ensemble. CCM-DES first divided feature space into different regions, and then constructed optimal ensembles in every region. It used DES for classify the unlabeled sample at last. Experiments on UCI datasets show that the criterion based on margin classifiers competence is better than current ordering criterion. Furthermore, CCM-DES has the advantages of smal- ler ensembles, higher accuracy, shorter classifying time than current ensemble algorithm.
出处
《计算机应用研究》
CSCD
北大核心
2015年第6期1698-1702,共5页
Application Research of Computers
基金
国家自然科学基金资助项目(61004069)
安徽省自然科学基金资助项目(1208085QF107)
