Öznitelik Mühendisliği ile Makine Öğrenmesi Yöntemleri Kullanılarak BIST 100 Endeksi Değişiminin Tahminine Yönelik Bir Yaklaşım

Abstract

: The main output of financial markets is a time series problem and a time series exhibit noisy, non-linear and chaotic structure by nature. Due to this complex structure, the process of predicting the future behavior of time series is a very challenging field for researchers. In this study, a comprehensive feature engineering process was applied to estimate the daily return direction of the BIST 100 index and models were carried out using different machine learning algorithms. The features to be taken as input to the models were extracted depending on the summative statistics of the series, the additional characteristics of the sampling distribution, and the observed dynamics reflecting the non-linear/complex structure of the series and it was shown that the classification performances are quite high without using exogenous variables. In addition the durability of the predictions performances was investigated using different training-test ratios.