The article is devoted to the consideration of topical issues related to the study of the possibility of forecasting the dynamics of stock markets based on neural network models of machine learning. The prospects of applying the neural network approach to building investment forecasts are highlighted. To solve the problem of predicting the dynamics of changes in the value of securities, the problems of training a model on data presented in the form of time series are considered and an approach to the transformation of training data is considered. The method of recursive exclusion of features is described, which is used to identify the most significant parameters that affect price changes in the stock market. An experimental comparison of a number of neural networks was carried out in order to identify the most effective approach to solving the problem of forecasting market dynamics. As a separate example, the implementation of regression based on a radial-basis neural network was considered and an assessment of the quality of the model was presented.
Keywords: stock market, forecast, daily slice, shares, neural network, machine learning, activation function, radial basis function, cross-validation, time series