LR, C5.0, CART, DVM yöntemlerini kullanarak hisse senedi getiri sınıflandırma tahmini yapılması ve kullanılan yöntemlerin karşılaştırılması: Türkiye’de BIST’de bir uygulama

Abstract

Predicting stock return classification is a research field that has always attracted the attention of investors and analysts. In this study, the factors that affected the stock returns were determined for 18 companies which were dealt at BIST 100 index, and operating to manufacture chemistry, rubber and plastic products, and were active between 2009 and 2014, after which their stock returns were predicted. Of all the data mining methods, relevant data were collected, and analyzed through LR analysis, C5.0 algorithm, CART algorithm and SVM methods. Accordingly, some rules were obtained out of the decision tree in order to reveal significant and beneficial information for predicting the stock return classification. As a result of the analyses, LR analysis showed 75% of success, C5.0 algorithm showed 88% of success, CART algorithm showed 89.8% of success, and SVM analysis showed 75.9% of success for accurate classification. It was determined that the most important variables influencing positive and negative predictions of the stock return classification were 'market/book value variable', 'CPI variable' and 'gross profit margin variable'. Using the studied model and its variables for predicting stock return classification was observed to be convenient and was suggested for investors and analysts.