Predicting of stock price movements with data mining techniques on Borsa Istanbul equity markets

In the field of finance with the developments of information technologies, data mining/machine learning techniques had rapidly taken their place. Both domestic and foreign literature review, it has been seen that the studies mainly include direction estimation of indexes. From this point of view, the aim of the study is to compare the correct classification rates of the direction predictions via including the stocks other than the recognized stocks to be made by data mining methods among the stocks traded in different markets of Borsa Istanbul Equity Market. The time interval (2430 days) between the date of 03 May 2010 and 31 December 2019 were selected as the working period and worked with daily and weekly periods. Stocks, BIST 100 Index (XU100), BIST All Shares-100 Index (XTUMY) and USD/TRY Banknote Selling rate and technical indicators that calculated with the help of these datas were determined as independent variables, then the direction of closing prices of the stocks was tried to be predicted. While preparing the prediction models, different data normalization (min-max [0, +1] and z-score) and feature reduction methods (Principal Component Analysis and Random Forests) were also compared. Logistic Regression, Support Vector Machines, Naive Bayes and K–Nearest Neighbor methods were used for predictions. The results show that, BIST Stars Group 1 stocks have greater accuracy rates than BIST Main Group 2 stocks. It has been difficult to come up with a clear result in terms of periods. Although the arithmetic average of the highest accuracy rates is obtained with weekly data, the arithmetic average accuracy rates of the models that prepared with daily data are higher mostly. It was seen that there was no difference between normalization methods and Random Forests are more successful feature selection method than Principal Component Analysis. The most successful prediction method is Support Vector Machines.