Research and Application of Deep Neural Network Architectures for Classification on Multidimensional Time Series

The results of studying the architectures of deep neural networks designed to solve classification problems are presented. As a result, attributes are formed for effective decision-making automation. Multidimensional time series of financial markets are used as data. The problems of binary and multiple classification are considered. Fully connected, recurrent (long short-term memory (LSTM)) and hybrid combined architectures of neural networks are analyzed. The studied multivariate time series is obtained by combining a one-dimensional time series of asset value, trading volume, technical indicators, and other parameters.