Abstract:
The bank loan is the most essential product that banks and other financial institutions offers.One of the most essential factors for banks to consider when deciding whether or not to grant a bank loan is the borrower's capability to repay the loan. As the banking sector improves, more people are looking for bank loans. However, banks have limited assets and can only offer loans to a certain number of people, so assessing who can get a loan and who will be a safer option for the bank is a common process. This research uses a mixed machine learning approach to Predict loan repayment capability of Oromia international bank customers. The researcher has collected 4411 loan dataset in Excel format from Oromia international. The dataset contains loan data including loan attributes from this bank from Dec,2015 to Jan,01/2022.
About 3598 (80%) of the dataset was used for the training model, while about 813 (20%) was used for the testing model. In This study four machine learning algorithms are used to predict loan repayment capability of customers: namely, random forest (RF), Decision Tree (DT), Support vector machine (SVM), and logistic regression (LR). According to the model performance results, the random forest (RF) outperformed and had the best prediction accuracy of 0.99%, while the DT, SVM, and LR had accuracy of 0.991%, 0.67%, and 0.75%, respectively, and the prediction accuracy after all models were combined/hybrid was 0.99.5%, indicating that the prediction results obtained from the combined/hybrid model will be useful in decision making for the bank who has the data. However, it is present because the purpose of this study is to predict customer loan repayment capabilities at Oromia International Bank. This thesis used a hybrid machine learning technique to detect trends in low risk and high risk or non-performing loan patterns utilizing historical data and build a predictive model to help the bank's management. Furthermore, because there are so many research works in this area, it is constructing that by comparing the suggested work to current works; it would be possible to demonstrate improvement in the prediction model
