Doctoral candidate Aurore Archimbaud from the University of Toulouse will give a presentation on ”Comparison of statistical methods for multivariate outlier detection” on Thursday 27.8., at 13:00, in room 284 (Quantum, 2 floor)

Abstract:
Detection of multivariate outliers is a relevant topic in many fields such as fraud detection or manufacturing defects detection. Several non-supervised multivariate methods exist and some are based on robust and non-robust covariance matrices estimators such as the Mahalanobis distance (MD) and its robust version (RD), the robust Principal Component Analysis (PCA) with its diagnostic plot and the Invariant Coordinate Selection (ICS).

The objective of this presentation is to compare these three different methods. Note that the three methods lead to one or several scores for each observation and that high scores are associated with potential outliers. The comparison is performed on simulated data sets with mixtures of gaussian distributions in the context of a small proportion of outliers and when the number of observations is at least five times the number of variables. While all variables are taken into account in the definition of the Mahalanobis distance, some components are selected for robust PCA and ICS. Our results illustrate that in case of a large number of variables, ICS is the only method that selects relevant components for outlier detection.

Doctoral candidate Aurore Archimbaud from the University of Toulouse will give a presentation on ”Comparison of statistical methods for multivariate outlier detection” on Thursday 27.8., at 13:00, in room 284 (Quantum, 2 floor)

Abstract:
Detection of multivariate outliers is a relevant topic in many fields such as fraud detection or manufacturing defects detection. Several non-supervised multivariate methods exist and some are based on robust and non-robust covariance matrices estimators such as the Mahalanobis distance (MD) and its robust version (RD), the robust Principal Component Analysis (PCA) with its diagnostic plot and the Invariant Coordinate Selection (ICS).

The objective of this presentation is to compare these three different methods. Note that the three methods lead to one or several scores for each observation and that high scores are associated with potential outliers. The comparison is performed on simulated data sets with mixtures of gaussian distributions in the context of a small proportion of outliers and when the number of observations is at least five times the number of variables. While all variables are taken into account in the definition of the Mahalanobis distance, some components are selected for robust PCA and ICS. Our results illustrate that in case of a large number of variables, ICS is the only method that selects relevant components for outlier detection.