Stocks Prices Prediction with Long Short-Term Memory

Abstract

It is a difficult problem to predict the one-day next closing price of stocks since there are many factors affecting stock prices. In this study, by using data from November 29, 2010 to November 27, 2019 and stocks for the closing price of the next day are predicted. The long short-term memory method, a type of recurrent neural networks, is preferred to develop the prediction model. The set of input variables created for the proposed model consists of stock price data, 29 technicals and four basic indicators. After the set of input variables is created, the one-day next closing prices of AKBNK and GARAN stocks are developed the model to predict. The model's prediction performance is evaluated with Root Mean Square Error(RMSE) metric. This value is calculated as 0.482 and 0.242 for GARAN and AKBNK stocks respectively. According to the results, the predictions realized with the set of input variables produced are sufficiently successful.