对不平衡类分类的一种组合方法  

Ensemble Approach for Class-imbalanced Classification

作　　者：冯力力[1] 李跃波[1] 苏宇[1] 王丽珍[1] 

机构地区：[1]云南大学信息学院计算机科学与工程系,云南昆明650091

出　　处：《广西师范大学学报（自然科学版）》2007年第4期277-280,共4页Journal of Guangxi Normal University:Natural Science Edition

基　　金：国家自然科学基金资助项目(60463004)

摘　　要：具有不平衡类分布的数据集在许多实际应用中是很常见的,但由于类分布不平衡,给那些已有的分类算法带来了很多问题。一种为处理不平衡类问题而开发的基于决策树和人工神经网络的有效组合方法被讨论。它基于数据抽样的方法构建组合分类器,并利用ROC曲线(Receiver Operating Characteristic curve)作为评价挖掘性能的分析工具,最后在PAKDD2007竞赛活动提供的实际数据上进行了有效性验证。It＇s common in application for data set with class-imbalance. Since the distribution of classes is imbalanced, there are some problems and challenges for current classification algorithms. In this paper, the thought of ensemble approach of decision tree and artificial neural network for class-imbalance is proposed. The method of ensemble classifier is based on data sampling. Performance of this method is evaluated with ROC （receiver operating characteristic curve）. Effectiveness is examined on the realworld data provided by PAKDD2007 competition.

关 键 词：不平衡类 决策树 人工神经网络 ROC 组合方法 

分 类 号：TP392[自动化与计算机技术—计算机应用技术]