Parallel Computation of Modified Stahel-Donoho Estimators for Multivariate Outlier Detection

Modified Stahel-Donoho (MSD) estimators are an orthogonally equivariant multivariate outlier detection method with a high breakdown point for all dimensions. An R function of the MSD estimators is created and its performance is confirmed, however, the method suffers from the curse of dimensionality and its implementation is limited to relatively low dimensional datasets. This paper proposes a parallel computing approach to cope with higher dimensionality and presents results for a few datasets to illustrate its use. Code for both the utilized parallelized function and the original single-core function have been placed in a public repository for further evaluation.