Searching for contaminants
N. T. Longford
Abstract
Decision theory is applied to the problem of identifying a small fraction
of observations that contaminate a random sample from a specified
distribution. The uncertainty about the parameters that characterise
the contamination is addressed by sensitivity analysis.
The analyst's (or the client's) perspective and priorities
are incorporated in the analysis by ranges of plausible loss functions.
An application to fraud detection is presented.