A METHOD TO SELECT DIAGNOSTIC-CRITERIA FROM BAYES THEOREM AND INFORMATION-THEORY

Abstract

An algorithm is proposed to order a list of characteristics according to their efficiency for diagnosis. Two diagnostic ability coefficients (DAC), INF [identification] and TAX [taxonomy], are described. The calculus of the INF coefficient is based on Bayes'' theorem, using the occurrence of various diagnoses, and on the information theory. The TAX coefficient is a linear approach of the former. Both coefficients are tested; 1891 strains of Enterobacteriaceae, originating from (human) urines, are studied. The results are compared with those given by Gyllenberg''s separation and characterization figures. Theoretical examples are provided for computing purposes.