← All research spotlights

Research spotlight · Unit 2

Peter J. Rousseeuw

KU Leuven, Section of Statistics and Data Science

Research field
Robust statistics, outlier and anomaly detection, cluster analysis

Peter Rousseeuw has spent four decades on robust statistics, the branch of the subject that asks how an estimate can be computed so that a minority of wrong or unusual values cannot drive it, and he developed several of its standard tools, among them least trimmed squares regression and the minimum covariance determinant estimator. In a survey with Mia Hubert he turns the idea around and uses robust fits to find the anomalies: once the majority of the data is described well, the cases that lie far from that description stand out. The survey starts from the median and the median absolute deviation, the robust summaries of this unit, explains why the mean and the standard deviation can hide outliers (masking) or make good values look suspicious (swamping), and extends the idea to data with many variables and to tables in which single cells rather than whole rows are wrong.

In the masterclass

Section 7, after the breakdown values of the mean, the median, and the MAD, shows how his robust summaries are used to flag anomalous observations.

The research is presented in Unit 2: Describing Distributions, in a research spotlight box that explains the result at the level of the unit and links it to what you have just learned.

The paper

Rousseeuw, P. J., and Hubert, M. (2018). Anomaly detection by robust statistics. WIREs Data Mining and Knowledge Discovery, 8(2), e1236. https://doi.org/10.1002/widm.1236

The scientists named here are not contributors to the Statistics Masterclass, are not affiliated with it, and have not endorsed it. The program presents and explains their published research, with references to the original papers.