Comparison of Random Forest and XGBOOST implementation's success to predict bank profitability: Evidence from Turkish deposit banks

Machine learning techniques are used as regression, classification, clustering or association rule inference. Therefore, one of the most important application areas of machine learning is classification. In this study, it is aimed to examine the factors affecting the profitability of banks. For this, 27 banks operating in Türkiye between the years 2010 and 2020 constitute the scope of the study. In the study, target variables return on assets (ROA) and return on equity (ROE) were used. Input variables are capital adequacy, asset quality, liquidity, management efficiency, and bank size. This study is aimed to examine the factors affecting the profitability of banks with Random Forest and XGBOOST algorithms along with Multi Linear Regression. Methods were compared according to error and explanation levels for each target variable. Then, the factors affecting the target variable were examined with the "target shuffling" technique. As a result, it has been found that management efficiency (activity), bank size, and capital structure are the most significant for ROA; while management efficiency and size are the most effective for ROE. JEL Classification Numbers: C38–G00 Keywords: Classification, Bank Profitability, Random Forest, XGBOOST