Credit risk management and credit risk measurement in the Turkish banking sector

Analyzing bank failures accurately is vital for both the local and global economy. The effects of bankrupt banks on the financial system deeply affect not only depositors and bank employees, but also other stakeholders of the financial system. Identifying and preventing potential crises and their possible effects early provides an advantage in terms of making financial stability sustainable. In this study, 6 different machine learning algorithms were used to classify bank failures in Turkey, and the sample of the study consists of financial data of 37 private commercial banks operating in Turkey between 1998 and 2000. To continue this research, after a series of preliminary experiments to find the best classifiers, Logistic regression, Multilayer Perceptron (MLP), Random Forest, J48, XGBoost (Gradient Boosting), GLMBoost (General Linear Model) models were used. After modeling, it was seen that the model with the highest prediction success was the Multi-Layered Perceptron model with 97% success. On the other hand, the study using the general linear model and Multilayer Perceptron appears as the models with the highest AUC value. It can be seen that both models have high ability to distinguish classes.