Stock Market Forecasting Using Machine Learning Models

Stock market forecasting is a challenging problem. In order to cope with this problem, various techniques and methods have been proposed. In this study, the stock close values are tried to be forecasted as monthly and weekly. For this purpose, values of two traded stocks (Google, Amazon) are predicted using with models being processed. In addition, stock data were taken from two different indexes (Dow Jones Industrial Average (DJIA) and S&P 500) for a realistic assumption. Stock market data includes long term dependencies. For this reason, classical Recurrent Neural Networks (RNN)-based models could not effectively work for in such data. Therefore, Long Short-Term Memory (LSTM) Networks and Gated Recurrent Unit (GRU) based models were developed and their efficiencies were observed in this study. These models were also compared with traditional machine learning approaches and the obtained gains were calculated. With LSTM, the close prices of datasets were predicted more consistently than the other models. In particular, the LSTM model is more successful than the GRU model with similar error metrics in dealing with fluctuations in datasets. Besides, linear machine learning models are well behind the deep learning-based models in weekly prediction of datasets.