Predicting the Credit Risk Classification of Retail clients at Rabobank

This thesis focuses on developing a prediction model to classify prospective credit risk for retail clients at Rabobank. Using machine learning techniques and a dataset from Rabobank's early warning credit risk monitoring system, the study explores various models, including Random Forest, XGBoost, SVM, NN, MLR, and LDA. Struggling to predict the minority class (CRC [Default]), the problem is restructured into a 3-class classification. An ensemble method combining Random Forest, XGBoost, SVM, and MLR emerges as the optimal model. Evaluation metrics indicate promising results, with potential cost savings in practical implementation. A proof of concept demonstrates the model's feasibility for real-world application, enhancing risk monitoring and mitigation strategies in the banking industry. The thesis offers valuable insights into credit risk analysis, with implications for improving credit risk management by contributing to minotoring with a forward looking approach.