Skewness and the Linear Discriminant Function

Fisher’s linear discriminant function might be difficult to estimate, when data come from a semiparametric, finite mixture model. We propose an estimator based on the singular value decomposition of the third standardized cumulant. The estimator is consistent when sampling from a mixture of two symmetric, homoscedastic components with finite third moments and different weights. We also evaluate its performance and compare it with another estimator, which uses the eigenvectors of a kurtosis matrix.