基于投影深度的判别分析

Discriminant Analysis Based on Projection Depth

用稳健的投影深度中位数代替样本均值,改进了多元正态分布下Bayes线性和非线性判别法;把投影深度引入到最大深度判别理论中,得到了基于投影深度的最大深度判别法,这种判别法对判别函数的形式和总体的概率分布都没有任何要求.模拟研究表明:新的判别方法相对Bayes线性和非线性判别法有着更强的稳健性.

The paper improved Bayes linear and nonlinear discriminant analysis through replacing sample mean and scatter matrix by robust estimates based on the projection depth. It also used the projection depth in the maximum depth classification method, and the classifiers do not assume any specific parametric form of the separating surface nor do they assume any particular type of probability distribution for the populations. The results of simulation studies showed that the depth - based classifiers are more robust than the Bayes linear and nonlinear discriminant analysis.