Using machine learning in assets to be selected for portfolio: An application on BIST participation 30 index

In this study, the aim is to predict the stock prices of companies listed in the BIST Participation 30 Index on the Borsa Istanbul using machine learning methods and to create various portfolio strategies based on the obtained prediction data. The study consists of three parts. The first part provides a theoretical framework for portfolio management. The second part introduces the concept of machine learning and its theoretical background. The third part of the study presents the analyses and findings conducted within the scope of the research and concludes with the results section. In the research, prediction results obtained using machine learning methods, specifically Linear Regression (LR) and Gated Recurrent Unit (GRU) algorithms from deep learning methods, are evaluated. The findings indicate that the Gated Recurrent Unit algorithm obtains prediction values with less error compared to the Linear Regression method. Based on the obtained prediction data, two different portfolios were created, one with equal weights and the other with return-based weights, using Markowitz's mean-variance model, and their performances were compared with the Borsa İstanbul (BIST) Participation 30 Index. The study concludes that portfolios created based on machine learning predictions exhibit high performance during the examined period. The results demonstrate that investors can effectively diversify and optimize their investment strategies by leveraging machine learning-based prediction models and portfolios constructed according to Markowitz model.