PREDICTING THE LOAN DEFAULT USING MACHINE LEARNING ALGORITHMS: A CASE STUDY IN INDIA

Abstract

The main income of banks was generated from mobilizing the deposits to borrow to applicants. Although applying for loans is becoming common, banks still need to take the risk that the applicants may have a loan default. In this study, the objectives are to predict the risk of loan default using 6 types of machine learning (Logistic Regression, Decision Tree, Random Forest, K-Nearest Neighbor, Support Vector Machine, and Naïve Bates), compare the machine learning algorithms to choose the most suitable algorithms for predicting the risk of loan default, and help the decision maker in approving or rejecting the loan requests. The dataset is focused on India with their behaviour to determine their risk. Using the Jupyter Notebook (Python) to build the model and evaluate each model. There are 5 types of evaluation metrics (Accuracy, Precision, Recall, F1-Score and Average) are used to determine the champion model among the six machine learning algorithms. In this study, K-Nearest Neighbor is the champion model because this model scored the highest in all the evaluation metrics, which is 0.89. Although machine learning algorithms can help to determine the risk of flagging, the decision maker should take some actions to decrease the risk of loan default such as creating a clear plan for the payment reminders and providing a convenient way.