分类器评估方法A-ROCCH的研究与实现  

Research and Implementation of A-ROCCH Evaluation Method for Classifier Performance

作　　者：邹洪侠[1] 陶硕[1] 

机构地区：[1]马鞍山职业技术学院电子信息系,安徽马鞍山243031

出　　处：《新乡学院学报》2017年第12期35-38,46,共5页Journal of Xinxiang University

摘　　要：以ROCCH为理论基础,根据多类转二类思想,采用ECOC完全编码,提出了一种代价敏感的三类分类器性能评估方法 A-ROCCH。在MBNC平台上对该方法进行了实现,与基于ECOC的AUC评估方法的实验数据作比较,证明了AROCCH方法的有效性、敏感性和健壮性。According to the idea of multi-classifier transforming to binary classifier,a cost-sensitive three-classifier evaluation method based on the theory of ROCCH was made up by using ECOC complete coding.The method was implemented on the MBNC platform;the experimental data was compared with the AUC evaluation method based on ECOC.The effectiveness,sensitivity and robustness of the A-ROCCH method were proved.

关 键 词：多类分类器 性能评佑 代价敏感 ECOC 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]