面向不平衡数据集的线性分类方法研究

Linear Classification Methods for Imbalanced Datasets

近年来,面向不平衡数据集的分类器学习与推广问题越来越受到人们的关注,在此以机器学习数据库、美国邮政编码、2维元音等国际上典型的分类问题为应用背景,重点研究如何用线性分类器解决样本数不平衡的问题;对Fisher、伪逆和单层感知器等3种典型的线性分类器做了深入的研究,并将这3种线性分类方法应用到不平衡数据集的分类中;通过实验及分析,这些新方法对平衡数据集的线性分类起到了良好的分类效果。

In recent years,much attention is paid to the learning and generalization problems of classifiers for imbalanced datasets.For the typical classification applications such as machine learning datasets,the US postal service,and 2-dimensional vowels,this paper focuses on the design and learning algorithms of linear classifiers in order to tackle the imbalanced datasets and makes deep studies on Fisher,Pseudo-inverse and single-layer perceptrons and applies these three linear classifiers to imbalanced datasets.Through experiments and analysis,these new methods play a good classification role in linear classification of imbalance datasets.