Outlier Detection and Minimum Distance Applications
Maximum likelihood procedures are widely applicable but are highly
influenced by the presence of outliers and bad data. In this
minisymposium, we survey a range of attacks on this problem.
One approach is to modify the influence function in order to
improve upon the performance of M-estimators. A second approach
is to modify the likelihood and incorporate a Bayesian framework.
Finally, the minimum distance approach may be adopted in a
semiparametric (eg. mixture of normals model) or nonparametric
(penalized minimum distance) framework. Modern data analysis
and data mining requires a range of outlier detection and
flexible modeling strategies.
Bill Szewczyk From Kernels to Mixtures: Modeling High-Dimensional Data
David Rocke Outlier Detection With Massive Data Sets
George Terrell Penalized Minimum Distance Modeling for Data Analysis
Will Wojciechowski A Bayesian MCMC Multivariate Outlier Detection Algorithm