A Novel Multivariate Recurrent Neural Network based Analysis Model for Predicting Stock Prices

Stock prices are a challenging issue in the real world because of their complicated dynamics and dependence on several factors. Complex time series issues have been successfully handled by recurrent neural networks (RNN). We examined several multivariate models built using various RNN architectures, such as GRU and LSTM, and compared them to their corresponding univariate models as well as to one another. This paper’s goal is to examine how well a neural network can deal with this situation. We have suggested a multivariate RNN-based analytical model in this work to forecast stock price movement. To determine the temporal relationships and anticipate stock prices of a certain firm from its prior historical data of stocks, we employed a soft computing technique based on RNN. In order to forecast stock price, we present a multivariate neural network structure and contrast and compare the estimation error of mentioned networks with a univariate LSTM model. The results highlighted that, when trained on the same data, the proposed multivariate prediction model readily beat the univariate prediction models. We have also demonstrated that multivariate prediction models may be trained and used in corporate environments more quickly and efficiently. The results suggest that the proposed RNN model gives the most accurate predicted result for the prediction with accuracy of 88.91% taking the time of 1.76 seconds. The accuracy for the LSTM and GRU is 85.43% and 81.24%. Time taken by GRU and LSTM ais 1.78 and 1.80 seconds respectively.