Application of Machine Learning for Predicting Stock Prices

In this paper we study the possibility and effectiveness of using classical machine learning methods to predict the trend of stock price movement. “Classical” in this context means those machine learning models that do not use neural networks. This problem seems to be relevant, as accurate forecasting on the stock market is of interest to both the scientific and professional community. In addition, many researchers believe that computer algorithms are the most promising tool for building such forecasts. Speaking of classical methods, it is worth noting that the concentration on them adds novelty to this work, since the vast majority of researchers build models based on neural networks, some of the classifiers we used were not used for this task before. Prior to direct modeling, the scientific literature was analyzed, which partially confirms the hypothesis that classical machine learning models can be accurate. Based on the literature studied, several classification models were selected and constructed. For this purpose, data on Sberbank stock quotes were collected and processed in the period from April 2005 to April 2019. Additional features were also generated from the primary data. As expected by the author, the most accurate algorithm was the boosting algorithm over the decision trees, which gave an accuracy of 62%. This indicator allows us to conclude that the model is effective and able to make predictions at a level or even better than the models of other researchers.