Predicting Bank Customers' Default Risk Using Machine Learning Algorithms: Emphasis on the Stacking Model

In this study, the main objective is to estimate and predict the probability of bank customers' default, rather than their credit rating. Accordingly, a set of machine learning algorithms was employed in the form of base models, including K-Nearest Neighbors, Naïve Bayes, Support Vector Machine, Random Forest, and Bagging Classifier. To evaluate the performance of each of these models, test data consisting of samples with different frequencies and a set including approximately ۱,۰۰۰ samples were analyzed. Evaluation metrics included accuracy rate and the area under the curve. The results showed that the stacking model, constructed by combining the results of the base models, outperformed the individual models. This model was able to achieve a prediction accuracy of default probability up to ۷۷%, with its ROC curve area reaching approximately ۰.۷۷, indicating better performance compared to other models in classifying customers at risk of default. The findings of this research emphasize that the use of ensemble approaches and stacking models in banking and financial systems can enhance the accuracy of credit risk assessment and serve as an effective tool in managing risk levels and reducing prediction errors. These results demonstrate that the more comprehensive and up to date application of machine learning techniques can provide significant support in banking decision-making processes and in reducing fluctuations related to loan defaults.

کلیدواژه ها:

Default Prediction ، Machine Learning ، Stacking Model ، Classification Algorithms