Abstract:
Financial institutions play an important part in offering enterprises and individuals’ opportunity in accessing affordable and suitable financial services. The goal of credit scoring is to discriminate against the bad applicants (defaulting) from those that are good applicants (repaying) as this is very important since it can save the financial institutions the losses. This thesis investigates the use of various machine learning models such as Linear Discriminant Analysis, Gradient Boosting,
Decision Tree, K Nearest Neighbors, Logistic Regression, Random Forest, Support Vector
Machine and XGB for automated credit scoring prediction. It has concentrated on comparison of performance of eight selected machine learning algorithms, establishing the best classifier for credit scoring prediction and finding the key determinants for credit scoring prediction. Credit data from one of Rwanda’s commercial bank was used in training and credit scoring models development. The performance of the algorithms was measured using various criteria such as accuracy, precision, f1-value, recall and AUC. Although the other models performed well, the results demonstrated that XGBoost outperformed the other algorithms showing better results with recall score of 0.9939, F1 score of 0.9892 and 99.56% accuracy hence it is considered the best algorithm for automated credit scoring on this dataset.
