Asymptotic results for expected probability of misclassifications in linear discriminant analysis with repeated measurements

In this paper, we propose approximations for the misclassification probabilities in linear discriminant analysis when the group means have a bilinear regression structure. First, we give a unified location and scale mixture expression of the standard normal distribution for the linear discriminant function. Then, the estimated approximations of misclassification are obtained for the three cases: unweighted case, weighted known covariance matrix Σ, and weighted unknown Σ. It has to be pointed out that larger p is better for classification when Σ is known, also in unweighted case. In the case Σ is unknown, we gain more information if fewer repeated measurements are used compared to when many repeated measurements closer to the number of included sample size are used. Furthermore, the accuracies of the proposed approximations are checked numerically by conducting a Monte Carlo simulation.