Cross‐Validatory Selection of Binary Variables in Differential Diagnosis

SUMMARY The statistical method of cross-validatory choice is used to construct a basis for prediction of a binary dependent variable from binary independent variables. The implementation of the method involves a score function, a cut-off value and a loss function. It proceeds by unrestricted sequential splitting, followed by a cross-validatory selection of the cut-off value which is then applied to the result of the splitting. The method is evaluated on three well-known data-bases for the diagnosis of kerato conjunctivitis sicca, rheumatoid arthritis and the treatment of jaundice. Generalizations of the technique are outlined.