Evaluation of BIST100 Index Prediction Performance of Deep and Machine Learning Algorithms

Abstract

This study investigates the possibility of forecasting the Borsa Istanbul BIST 100 index using machine learning and deep learning techniques. The study uses the BIST 100 index as the dependent variable. In addition, gram gold price, daily dollar exchange rate (in TL), daily euro exchange rate (in TL), BIST trading volume, daily Brent oil prices, BIST trading volume, BIST overnight repo rates, and BIST Industrial Index (XUSIN) data are used as independent variables. The Central Bank of the Republic of Turkey provides daily statistics on these variables. The performance of several deep learning recurrent neural networks (RNN) and machine learning network structures—including Random Forest, K-Nearest Neighbors, Multilayer Perceptron, Radial Basis Function, and Support Vector Machine—for predicting the BIST 100 index is tested and compared in this study. The results indicate that the CNN model outperforms the other models in terms of prediction accuracy, with the lowest RMSE and MSE values, and the highest R² value. This suggests that CNN is a robust model for financial forecasting. The relevant literature is summarized in this context in the first portion of the study, after which the methods and results are described. Then the obtained comparative prediction values are presented. Finally, the study is concluded by presenting the interpretations of the results and recommendations.