摘要
传统Fisher判别方法不具稳健性,其分析结果易受异常值的干扰。本文采用近年新兴的投影追踪(Projection-Pursuit)的思想,将高维数据降维而估计出稳健的协差阵,进而建立一种新的能抗异常值干扰的稳健的Fisher判别方法。对该方法进行了计算机Monte Carlo模拟比较,结果显示当异常值存在时,该法优于传统的Fisher判别方法。
Fisher Discrimination is a important method in the aera of multivariate analysis, and it has being widely used in biomedicine, geology, economics and other scientific aeras. But. it is nonrobust and can easily be influenced by outliers. In this paper, we developed a new robust Fisher discrimination procedure via the technique of Projection-Pursuit. Results of Monte Carlo simulation and practical examples showed that the new procedure wins advantages over the classical Fisher discrimination when some outliers exist in data set.
出处
《数理医药学杂志》
1993年第3期21-26,共6页
Journal of Mathematical Medicine
基金
国家自然科学基金(编号 39070765)
关键词
判别分析
稳健统计
Discriminant analysis
Projection-Pursuit
Robust statistics