Hybrid Cellwise-Robust Fisher Discriminant Analysis for High-Dimensional Data

Fisher discriminant analysis (FDA) is sensitive to outliers, and when the number of variables p exceeds the number of observations n, the sample covariance matrix is singular.We propose a hybrid robust covariance estimator that combines cellwise detection and imputation (Detect Deviating Cells) with rowwise MRCD estimation, mixes the two components with a weight, and applies ridge shrinkage so that the estimator is invertible for p > n. The estimator is used in a modified Fisher rule.