B1401
Title: Fast robust correlations with application to cellwise outliers
Authors: Peter Rousseeuw - KU Leuven (Belgium) [presenting]
Jakob Raymaekers - KU Leuven (Belgium)
Abstract: The Pearson (product-moment) correlation coefficient is sensitive to outliers in the data. The literature contains quite a few other correlation measures which are more robust to outliers. Many of these have a higher computational complexity than Pearson's O(n), especially if they are computed from very robust scatter matrices. But in high-dimensional data there are many variables to correlate, so speed is essential. We compare several robust correlation measures with a view toward computation time, preferably no more than O(n log(n)) for a pair of variables, and the ability to form positive semidefinite correlation matrices. We then apply the selected measures to multivariate data analysis techniques including cluster analysis, classification and the detection of cellwise outliers.
