A Combined Model of ARIMA-GRU to Forecast Stock Price

Abstract

Deep learning has gained more interest and attention in comparison with machine learning in recent years. It has been successfully applied to many applications of the real world to solve problems on a daily basis. Time series forecasting for the stock market is the most challenging area in the time series forecasting domain. In this paper, we propose the use of deep learning technique in stock price forecasting and compare the same with machine learning technique. We present hybrid models of auto-regressive integrated moving average support vector machine and auto-regressive integrated moving average gated recurrent unit, respectively. The closing stock prices act as inputs for the auto-regressive integrated moving average model that gives residuals of the same as output. The residuals along with the original closing stock prices are used to compute the new closing stock prices, that is, put to the deep learning and machine learning models, respectively, to predict the next day’s closing price. By experimentation, we are able to conclude that auto-regressive integrated moving average gated recurrent unit is able to give a better accuracy in comparison with auto-regressive integrated moving average support vector machine on long-term forecasting of the closing price of stock for a large data that includes three different datasets taken from the Indian National Stock Exchange.