E.N. Zguralskaya

Traditionally it was thought that using artificial neural networks (ANN) may well solve the problems of classification and prediction, but rather to explain the bad decision-making process. Refute this opinion was by ANN models with minimal configuration. The learning process in these models does not depend on predefined network configurations as first the synaptic weights, and then use them to construct a neural network. Configurable through a solution of the minimal covering the training set objects, standards. For modeling process of intuitive decision-making, the computation of sets of informative features of different types in order to determine the ANN algorithm with minimal complexity correctly (no errors) recognizes objects of training set. The complexity of classification algorithms is defined as an estimate of time spent on the decision to classify an arbitrary admissible object to one of disjoint classes. In the informative sets selected various types of attributes, correlation of each of which is minimal. Determined by the correlation of a square matrix with the values of the measures of closeness between the classes of all pair wise combinations of traits and contributions of each attribute in the decision making process of assigning objects to classes. To minimize the number of searches using the ordering features relative complexity of the algorithms of ANN with minimal configuration. In the computational experiment on data from two classes (patients with arterial hypertension and practically healthy peoples), the minimum informative set was presented to the 6th symptoms: arterial systolic blood pressure, mean arterial pressure, pulse pressure, arterial diastolic pressure, left arterial cavity size, age