A COMPARATIVE ANALYSIS OF THE RESULTS OF CONVENTIONAL AND COMBINED METHODS OF TRAINING DIRECT PROPAGATION NEURAL NETWORK IN HEALTHY PERSONS TO DETECTING THE DEGREE OF ACTIVITY OF AN AUTONOMOUS NERVOUS SYSTEM

The article deals with a comparative analysis of the efficiency of an artificial neural network (ANN) trained with the help of algorithm back propagation, and of that trained by combining a back propagation algorithm and a variant of Cauchy stochastic training, in detecting the degree of activity of an autonomous nervous system. For the purposes realization of the project has been developed a biotechnical system, including technical device for input electrophysiological information in mode on-line. To evaluate the clinical effectiveness of the classification, records of interpulse intervals in 139 healthy students of Belgorod State University have been analyzed. All of them were part of the same age and social group from 17 to 24 years old. In practice, the ANN training algorithm using the back error propagation method enabled a correct recognition of 96.0% of the samples. The analysis of the clinical effectiveness of combining the back error propagation algorithm with Cauchy stochastic training showed that 100% samples were detected correctly both: as in training statistical samples so and in examination sampling. Classification errors amounted to 0 %. Keywords: interpulse intervals; electrophysiological information input block; neurocomputing; neural network classification algorithm; method backward propagation algorithm; Cauchy stochastic training; combined methods of training.