Computer Engineering and Applications ›› 2012, Vol. 48 ›› Issue (5): 156-158.

• 数据库、信号与信息处理 • Previous Articles     Next Articles

Evaluation method of multi-class classification BO-AUC

QIN Feng, YANG Fan, CHENG Zekai, LIU Niu   

  1. School of Computer Science, Anhui University of Technology, Ma’anshan, Anhui 243002, China
  • Received:1900-01-01 Revised:1900-01-01 Online:2012-02-11 Published:2012-02-11

BO-AUC多类分类评估方法

秦 锋,杨 帆,程泽凯,刘 牛   

  1. 安徽工业大学 计算机学院,安徽 马鞍山 243002

Abstract: AUC assessment is the hot pot in data mining, in which B-AUC assessment can effectively assess the classifier performance, but the evaluation method has three disadvantages. This classification is based on two types of asymmetry, which impacts the evaluation results; by non-complete binary tree resulting in waste of storage space; the search based on partial binary tree is not efficient. This paper proposes BO-AUC assessment method which is based on complete binary tree structure and characteristics. This method can effectively solve the shortage of B-AUC. Further expansion of the standard, programmed algorithm is given. Experimental results show that the BO-AUC method achieves good results of classification.

Key words: Area Under roc Curve(AUC) evaluation, Binary-Area Under roc Curve(B-AUC), complete binary tree, Optimized Binary-Area Under roc Curve(BO-AUC), classifier performance

摘要: 分类技术是数据挖掘研究的核心技术之一,分类评估也是研究热点,基于AUC评估方法是分类评估领域的研究热点,其中B-AUC评估算法可以有效地评估分类器性能,但该评估方法有不足之处。该分类评估方法建立在不对称的两个类别上,影响了评价结果;根据非完全二叉树思想存储,浪费了存储空间;基于偏二叉树的搜索效率不高。利用完全二叉树的构造思想提出了BO-AUC评估方法,该方法将n个类别的分类问题分解为独立的二类进行成对的计算,可以有效地解决B-AUC的不足,进一步扩展基于AUC的评估标准,在MBNC实验上编程实现该方法,实验结果表明BO-AUC方法的有效性。

关键词: 曲线下的面积(AUC)评估, 基于二叉树方法求的曲线下的面积(B-AUC), 完全二叉树, 优化的基于二叉树方法求的曲线下的面积(BO-AUC), 分类器性能