Automatic classification of biomedical information when classification error is unknown.

Abstract

This paper describes a distribution-free method for the automatic classification of individuals by phenotype of the serum protein alpha 1-antitrypsin. This method achieved 74 per cent agreement and a Kappa value of 0.24 when compared to the conventional classification procedure on an independent testing sample. The methodology is appropriate for many classification problems, particularly when the true classes and the reliability of the conventional procedure are unknown.