Hybrid Intelligen System Based on an Adaptive Fuzzy Neural Network with a Data Input Search Block

The system is based on an adaptive neuro-fuzzy network. A method of initial attribute space reduction based on the Independent Component Analysis (ICA) is proposed. The models obtained on the basis of these methods have a linear form and are much simpler than the neural-fuzzy network, but the obvious drawback of the methods is the impossibility of mathematical describing the nonlinear relationships that are always present in real tasks, and the models obtained are not suitable for direct solution to intellectual tasks. In paper propose the hybrid intellectual systems (HIS) consisting of two blocks: a data input search block and a decision block. Data input search block is based on the ICA. The decision block is represents a cascade connection of adaptive neuro-fuzzy networks. Networks have the same structure and differ in the values of the coefficients that are determined during learning.