Oliynyk A. Evolutionary methods of feature selection and neural network model construction for pattern recognition

The thesis is devoted to a development and research of a feature selection and neural network model construction methods based on evolutionary approach. The new methods and software that used the available information about modelled object or dependence and based on the evolutionary approach are developed. These methods allow to perform feature selection, parametrical, structural and structural-parametrical synthesis of neural networks, and their simplification. The experimental research of properties and characteristics of the developed methods and software by neural network models synthesis have been carried out for solving real-world problems in technical and biomedical diagnostics. The proposed methods allow to reduce a search time, and neural networks, constructed with the help of the suggested methods, contain the most significant interneural connections.