The banking system, as a cornerstone of a country’s financial structure, plays a vital role in economic development. One of the key challenges banks face in offering loans is accurately identifying customers and distinguishing between different groups. This study aims to classify bank customers based on credit risk and predict their probability of default using machine learning models. The results of customer segmentation with the K-means algorithm indicate that customers are divided into four main clusters. The fourth cluster includes high-risk customers with significant overdue debts and defaulted loans, requiring close monitoring. The third cluster consists of crisis-prone customers with high overdue debts who are at risk of moving into the high-risk category. The second cluster represents medium-risk customers who, although relatively stable, need monitoring. The first cluster comprises low-risk customers with minimal overdue debts, making them ideal candidates for new loans. This research utilizes XGBoost and logistic regression models to predict defaults. The proposed hybrid model helps banks design more effective risk management strategies, optimize financial resource allocation, and strengthen their competitive position in the market.