On the Asymptotic Distribution of the Location Linear Discriminant Function
Journal of the Royal Statistical Society: Series B (Statistical Methodology) · 1985 · Volume 47 · Issue 3
DOI: 10.1111/j.2517-6161.1985.tb01380.x
Abstract
The location linear discriminant function forms the basis of the optimal Bayes classification rule when the data contain both binary and continuous variables (Krzanowski, 1975). When the unknown parameters are replaced by their sample estimates, the resulting “estimative” rule has an exact distribution which is too complicated for numerical use. We present here an asymptotic expansion of this distribution and study its performance under various assumptions about the parameters.
Keywords
discriminant analysis, mixture of continuous and binary variables, location model, actual and expected error rates, mahalanobis distance
Enable JavaScript for citation tools and additional paper interactions.