Derin Öğrenme Yöntemleri ile Borsada Fiyat Tahmini

Abstract

In recent years, due to the technological advances in computer hardware and enhancements in machine learning techniques, there are two hot research areas in problem solving, the use of "Big Data" and "Parallel Processing". Many real-world problems can be solved with the use of different Deep Learning algorithms, which can be realized in parallel with multicore computing devices such as GPUs. Deep learning models show great success in applications such as classification of raw data, regression analysis and estimation in time series. One of the most active application areas of these models is “the financial market” which aims a good estimation of stock prices in the exchange market. In this paper, it is aimed to forecast the short or long term future value by looking at the previous log data of the stock on the process of change in the market. In this study, a price forecasting system was developed by using 3 different deep learning models named LSTM, GRU and BLSTM with a comparative analysis between them. The time series values of the stock were used from the New York Stock Exchange from 1968 to 2018 as a set of data to be free of speculative movements. Specifically, tests were conducted on IBM stock. Experimental results show that the directional accuracy of 63.54% was achieved with the BLSTM model where the last 5-day transaction data of the stock were used as input.