A Deep Neural Network Architecture for Solving Classification Problems on Multidimensional Time Series of Financial Markets

The paper describes the results of a study of existing architectures of deep neural networks designed to solve classification problems. Multidimensional time series of financial markets are used as data. The problems of binary and multiclass classification are considered. Fully connected, recurrent (LSTM) and hybrid combined architectures of neural networks are analyzed as tools. The studied multidimensional time series is obtained by combining one-dimensional time series of the asset value, trading volume, technical indicators, as well as some other parameters.