Ryepka V. Neural network models for selection of estimation parameters methods for regression dependence in information control systems

The thesis deals with development of neural networks classification models that provide a choice of efficient parameters estimation method for mathematical models of processes with linear dependence between input information and influences of varios random factors to solve the determinate functional tasks in information control systems. The method for designing a composite neural network classifier is developed. The neural classification models for recognizing the classes of biased, robust and regression method estimation are obtained, the neural classification models for recognition of the subclasses of ridge regression methods are obtained. Generalized mathematical models that establish dependence between efficiency estimation criteria and statistical characteristics of the input information are developed.