Identification of Multivariate Outliers

Outliers, or unusually extreme values, have§traditionally been viewed as a nuisance to§researchers. Classical statistical analysis can lead§to completely opposite conclusions if outliers§are present or absent. Such points can, however,§alert the researcher to unexpected features hidden§within in a data set, and lead down paths of§surprising discovery. Outliers could even be the§primary purpose of the investigation. Credit card§fraud, electronic network intrusions, and§unusual stock characteristics preceding a large move,§for instance, can all be seen as outliers whose§presence is important to establish as quickly as§possible. Several methods have been§proposed to identify outliers, but many of these are§not computationally suitable for large data sets.§This book presents a review of multivariate outlier§identification with particular emphasis on large data§sets, and investigates a new method. The intended§audience is statistics practitioners and data§analysts who wish to detect outliers, as well as§those interested in the historical development of the§field. Basic familiarity with statistical concepts is§assumed.