摘要
主分量分类器是在最大化样本投影代数和的前提下求解分类面法方向,并采用核方法解决线性不可分问题.它对所有真实世界样本(包括野值)的重视程度相同,且只考虑了一个野值.设计并实现了一类鲁棒性较主分量分类器更强的增强型主分量分类器,其中重点讨论了三种典型权设置下的分类器特性.分析和实验证实了增强的主分量分类器的抗野值、噪声性能以及学习和推广能力均优于主分量分类器.
Principle Component Classifier (PCC) takes the normal vector of a hyperplane as the projecting direction, onto which the algebraic sum of all samples' projections is maximized, and deals with linearly nonseparable problem using kernel tricks. It is unreasonable in real world that PCC views outliers, together with noises, as important as other normal samples.A classifier named Enhanced Principle Component Classifier (EnPCC) with better robustness has been designed and implemented. Especially, the characters of the classifier with three typical weights are discussed. Finally, analysis and experiments on one toy problem and a benchmark dataset show the superiority of EnPCC to PCC in eliminating outliers or noise, learning ability and generalization.
出处
《复旦学报（自然科学版）》
CAS
CSCD
北大核心
2004年第5期769-772,共4页
Journal of Fudan University：Natural Science
基金
国家自然科学基金资助项目(60271017)
江苏省自然科学基金资助项目(BK2002092).
