Analyzing and scoring bank customers with machine learning

The banking sector supports economic growth by providing loans to businesses, and the accurate assessment of credit risk is a critical requirement for the sustainability of these processes. In Turkey, the diversity in the financial structures of businesses challenges the limitations of traditional credit scoring methods. This thesis aims to analyze and score bank customer firms using machine learning techniques. The dataset in this study includes information about the financial status, risk data, and the number of firms' employees. The study follows a two-phase analysis process. In the first phase, firms are grouped into four clusters based on risk data using the K-Means algorithm. The clusters obtained in the first phase are then used as independent variables in the model. In the second phase, supervised machine learning algorithms, including XGBoost, Decision Tree, Random Forest, KNN, and LightGBM, predict the credit score. The credit score is a four-class dependent variable (0-1-2-3). Model performance is evaluated based on criteria such as precision, recall, and F1 score, yielding successful results. This study aims to provide a unique contribution to the existing literature by segmenting firms using unsupervised machine learning methods in addition to considering their financial status and risk data and using the generated segmentation in credit scoring.