一种面向单个正例的Fisher线性判别分类方法  

A Fisher Linear Discriminant Classification Approach Dealing With Single Positive Sample

作　　者：尹军梅[1] 杨明[1] 

机构地区：[1]南京师范大学数学与计算机科学学院,江苏南京210097

出　　处：《南京师范大学学报（工程技术版）》2008年第3期61-65,共5页Journal of Nanjing Normal University(Engineering and Technology Edition)

基　　金：国家自然科学基金(40771163)资助项目

摘　　要：提出了一种解决不平衡数据集中少数类只有一个样本的方法,找出单个正例在负类中的k个近邻,按照一定规则依次在单个正例和它的各个近邻的连线上产生合成样本,并把这些合成样本添加到原始的正类中,用加权F isher线性分类方法对新的数据集进行训练.实验结果表明该方法可有效地提高少数类的分类性能.An approach to dealing with imbalanced data set with only one positive sample is proposed. After finding out the K-Near-Neighbours （K-NN） of the single positive sample, according to certain rules, synthetic samples are produced in turn on the connected lines between the single positive sample and every near neighbour of it. Then the produced synthetic samples are added to the original positive classes. Further, the new data set is trained with the weighing Fisher linear dis- criminant classification approach. In the experiment, eight data sets are chosen from UCI, and the data sets are trained. The results show that this approach can improve the classification performance of the minority classes effectively.

关 键 词：不平衡数据集 FISHER线性判别 过抽样 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]