Credit Risk Prediction Models Using Ensemble Neural Networks: The Case Of Oromia Credit And Saving Association

Abstract

Credit risks persist in Ethiopia's expanding economy. It is intolerable to overestimate the value of credit facilities. In today's global economic crisis, having a sound strategy for dealing with credit risks and offering a powerful and accurate model for credit risk prediction is crucial. The Oromia credit and saving association struggles in finding out the potential customers to study before lending loan to them. Credit money are risky in business process. They take a variety of financial risks in the course of providing financial services. Credit risk in Credit and saving association is usually defined as the possibility of a borrower defaulting his loan commitments. In Oromia credit and saving association probability of a loss arising from a borrower's letdown to repay a loan or meet contractual commitments is indicated as a credit risk. It regularly refers to the risk that the association was not obtain the payable principle and interest, resulting incurred in cash flow and greater collection costs. There is a high demand for predicting the credit risk as it is the needed one for analyzing loan losses and non-performing loans for decision makers in the financial institutions. The purpose of this research study is to build an ENN models to predict credit risk in the Oromia credit and saving association. Totally 20,808 credit transaction raw data were gathered from Oromia credit and saving association, which exported Excel data from their data base. The dataset contains credit transactions with 13 loan attributes from 2010 to 2021. The training model used 16,646 (80%) of the dataset, whereas the testing model used 4,162 (20%). Wavelet transformation was used for preprocessing. Deep neural network models extract the features needed for modeling. Neural network models like RNN, LSTM, and MLP were built for credit risk prediction. The ensemble neural network model (LSTM RNN-MLP) provides better accuracy of 99.4% which is better than the related works. The prediction results obtained from the model be helpful in plaining and management for credit risk reduction