MODELS AND METHODS OF INTELLECTUAL ANALYSIS FOR MEDICAL-SOCIOLOGICAL MONITORING’S DATA BASED ON THE NEURAL NETWORK WITH A COMPETITIVE LAYER

In this scientific publication, we suggest using the system of intellectual analysis of medical and sociological monitoring's data using a neural network with a competitive Kohonen layer to automate the process of obtaining knowledge (metadata) about the state of public health of the target audience. The following specialized tools have been developed to implement the system: models and a method for presenting detailed and aggregated medical and sociological data in area of primary and secondary features; the method of neural network classification of respondents based on machine learning of a neural network with a competitive layer; the procedure for labeling neurons of the Kohonen layer, taking into account the classification decisions received from the sociologist-analyst (initial markers). At at the first step, a two-dimensional histogram of pairwise coincidences of neuron numbers and existing initial class markers was constructed, and then it was corrected by lines and by columns in accordance with the developed rule. The result of the correction is the correspondence matrix of the numbers of neurons of the Kohonen layer and existing markers of classification decisions. The testing of the developed models and methods is based on a system of intellectual analysis using real medical-sociological monitoring's data. The research results show that it is possible to increase the relative share of correct classification decisions by an average of 20 % and reduce the share of false decisions by 50% compared with the sociologist-analyst for tasks of intellectual analysis of medical and sociological monitoring's data. These tasks were related to determining the working conditions of respondents.