Supervised machine learning for loan approval prediction model
Abstract
Loan approval is one of the most critical processes that any financial institution possesses. Any loan application's approval or denial has a direct impact on the bank's revenue and profitability as reported in quarterly financial statements. Though loan approval is a necessary step, the actual choice is not a simple one and is fraught with uncertainty. This research study aims to develop a loan approval model using a machine learning algorithm that can accurately predict loan approval for deserving customers in a speedy, fast, and simple way. A comprehensive analysis of eight different models, including Random Forest, Decision Trees, Support Vector Machine (SVM), Logistic Regression, Gradient Boosting Machine, Cat Boost, Deep Neural Networks, and XGBoost, was conducted. The evaluation of these models involved training accuracy, validation accuracy, F1 score, and the area under the receiver operating characteristic curve (AUC). The best model was chosen basing on the AUC and in particular, decision trees was the best model as it achieved an AUC of 0.70. Recommendations include full adoption of machine learning models especially decision trees to help banks in Uganda reduce the burden of manual loan approval and increase client satisfaction as well as revenues for the bank. In addition, five most important variables should be taken into consideration when lending to the customers namely; loan amount, credit history, applicant income, coapplicant income and loan repayment period.