Дослідження Модифікації Гібридної Рекурентної Нейронної Мережі для Прогнозування Цін Акцій

Dmytro Kachan; Nadezhda Nedashkovskaya

Research on Modification of a Hybrid Recurrent Neural Network For Stock Price Forecasting

Authors

Dmytro Kachan Institute for Applied System Analysis, Kyiv Polytechnic Institute
Nadezhda Nedashkovskaya Institute for Applied System Analysis, Kyiv Polytechnic Institute

Keywords:

recurrent and convolutional neural networks, LSTM, GRU, encoder-decoder, sequence-to-sequence models, attention mechanism, regularization

Abstract

A hybrid neural model is proposed, which combines recurrent, convolutional layers and an attention mechanism for stock price prediction. Experimentally, the model is compared with LSTM and GRU models of autoencoders. A system for forecasting stock price using statistical and machine learning methods has been developed