METHODS OF INTELLIGENT DATA ANALYSIS USING NEURAL NETWORKS IN DIAGNOSIS

The considered methods make it possible to develop the structure of diagnostic systems based on neural networks and implement decision support systems in classification diagnostic problems. The study uses general special methods of data mining and the principles of constructing an artificial intelligence system based on neural networks. The problems that arise when filling knowledge bases and training neural networks are highlighted. Methods for developing models of intelligent data processing for diagnostic purposes based on neural networks are proposed. The authors developed and verified an activation function for intermediate neural levels, which allows the use of weighting coefficients as probabilities of diagnostic processes and avoids the problem of local minima when using gradient descent methods. The authors identified special problems that may arise during the practical implementation of a decision support system and the development of knowledge bases. An original activation function for intermediate layers is proposed, obtained based on the modernization of the Gaussian error function. The experience of using the considered methods and models allows us to implement artificial intelligence diagnostic systems in various classification problems.