Principal component discriminant analysis.

The approach adopted involved two-stages. First the 11205 measurements in the mass spectrometry data were reduced to 14 scores by a principal component analysis of the centered but otherwise untreated and unscaled data matrix. Then a linear classifier was derived by linear discriminant analysis using these 14 scores as inputs. This number of scores was chosen by leave-one-out cross-validation on the training set, where it gave an overall error rate of 14%. Some indication of the information used in the classification may be obtained from an inspection of the coefficients of the linear classifier.