Detection of outliers with boxplots

Abstract

Low-complexity robust modifications to the Tukey boxplot based on fast highly efficient robust estimates of scale are proposed. The performance of the Tukey boxplot and its modified robust versions is measured relative to identi- fication of outliers in Monte Carlo experiments at contaminated normal distri- butions. The obtained results show that the proposed methods outperform the conventional Tukey boxplot and the classical Grubbs test.