Automatic Generation of Neural Network Topology to Forecast the International Integration Processes

The article is devoted to the automatic generation of neural network topology based on genetic algorithm to solve the problem of forecasting indicators of international integration processes. The paper proposes and implements a modification of the classic algorithm NEAT, consisting in the automatic selection of activation functions, which allowed to expand the set of possible solutions. The proposed method of predicting targets using the window method based on pre-predicted factor variables. This approach significantly increases the accuracy of forecasting.