Abstract— The approximation of the discriminant Anderson function at a point by a linear function in the feature space used in supervised learning allows us to estimate the posterior probabilities of classes at a point without using conditional probability distributions of feature classes and without using add-ons such as the Platt calibrator. We used the method of evaluating the posterior probability of a class to develop a computer program for differential diagnosis of oral lichen planus with oral leukoplakia and oral squamous cell carcinoma. Diagnosis by the posterior probabilities of diseases in the examined patient increases the reliability of the diagnosis and greatly facilitates the choice of treatment tactics.