engleski

Using machine learning tools in Nasdaq Composite stock market index prediction

In recent years an incredible progress has been made in the field of artificial intelligence, deep learning and machine learning. There are various applications of machine learning in economics today such as virtual personal assistants, predictions while commuting, videos surveillance, social media services, email spam and malware filtering, online customer support, etc. Machine learning or data mining techniques are used with some success for predicting stock prices and are gaining popularity in recent years. However, prediction of financial markets is a very difficult task because factors that influence stock price formation are complex and hard to predict. The main factors are political situation, expectations, economic conditions and unexpected events. In the process of predicting stock prices the most common approach is to use historical time-series data. This paper contributes to this important field of research. The aim of the paper is to use machine learning tools in Nasdaq Composite stock market index prediction. For that purpose an open source machine learning software Weka was used accompanied with its algorithms (Linear regression, Gaussian Processes, SMOreg and neural network Multilayer Perceptron). Multilayer Perceptron is a feed-forward artificial neural network composed of multiple layers or perceptrons. The prediction of close price of Nasdaq stock market index was made for 15 working days, in the period from the January 3rd to the January 24th 2020 using training data for the past five years. The dataset contained data for 7 attributes: open price, high price, low price, close price, adjusted close price and volume of transactions. The performance of aforementioned data mining techniques was evaluated using standard statistical error measures, MAE, MSE, MAPE and RMSE. Actual and forecasted values of Nasdaq stock market index are then compared with calculated difference between these two values. Machine learning technique that has the smallest errors can be accepted as the most accurate one or best suited to data. The results of the analysis have shown that vector machine SMOreg is the most accurate method for predicting the close prices of the Nasdaq stock market index. On the other side, neural network Multilayer Perceptron and algorithm Gaussian Processes were ranked lower on the precision scale. Limitation of the paper is related to the analysis of only one major stock market index so the results of the analysis should not be taken as final. Therefore, further investigation should be made in this field by analysing other stock market indexes or individual stock prices in order to get more conclusive results. Although a prediction of stock price movements is a very difficult task, finding a best prediction method can be helpful to investors when making decisions about their actual buy-sell strategy.

Machine learning, neural networks, WEKA, stock price prediction, Nasdaq Composite index

nije evidentirano