A comparative study of machine learning techniques for stock price prediction

Stock price prediction has garnered significant interest among researchers and investors. Machine learning has shown great potential to produce accurate forecasts in the past few years. This paper has applied several machine learning techniques to develop a valid forecast consisting of linear models and various artificial neural networks. We have tested our models on the daily EURUSD pair dataset from the foreign exchange market and the daily S&P 500 dataset from the US stock market. Lastly, we have generated a fair comparison between different models and defined best practices for each domain. Our results indicate the efficiency of the linear models on the EURUSD dataset. Moreover, although deep neural networks have the best performance in predicting the exact price of the S&P 500, we found out that the ARIMA model can forecast the direction of the stock price better than any other model.

2022 8th Iranian Conference on Signal Processing and Intelligent Systems (ICSPIS)