Classification performance comparison of artificial neural networks and support vector machines methods: an empirical study on predicting stock market index movement direction

Abstract

In this study, Artificial Neural Networks and Support Vector Machines which are widely used machine learning methods were examined. Usability of these methods for the prediction of the Istanbul Stock Exchange (ISE) National 100 Index (currently named BIST—100) movement direction was investigated. In the analysis, performances of these methods on the 2005–2011 period data sets containing technical indicators, other stock market indices and common macroeconomic indicators were compared. The results showed that technical variables give better performances than other variables. Later, a data set that predicts the stock index movement direction most accurately with a minimum number of variables was formed by feature selection on the aggregation of the mentioned data sets. Artificial Neural Networks gave better results than Support Vector Machines for all analyzes.