Development of a Bank Customers' Credit Rating System Using Back Propagation Neural Network and Feed-Forward Propagation Neural Network Algorithms

Abstract

Credit rating systems play a critical role in financial institutions by assessing customer creditworthiness, mitigating credit risk, and reducing financial losses from defaults. Traditional credit scoring methods primarily rely on historical financial data, which limits their adaptability in dynamic financial environments and excludes potential borrowers with limited credit histories. To address this gap, machine learning (ML) techniques have been increasingly adopted to enhance credit rating models. This study develops an advanced credit rating system leveraging Backpropagation Neural Networks (BPNN) and Feedforward Propagation Neural Networks (FPNN) to improve prediction accuracy and risk assessment. The models were trained using financial transaction and credit history data from Nigerian banks, incorporating both financial and non-financial indicators. BPNN was chosen for its ability to iteratively adjust neural network weights through backpropagation, while FPNN served as a baseline due to its simpler forward computation process. A systematic approach, including data collection, preprocessing, model training, and evaluation, was employed. Performance was assessed using key metrics such as accuracy, precision, recall, and F1-score. Results showed that BPNN outperformed FPNN, achieving an accuracy of 82.5% compared to 81.2%, with higher precision, recall, and F1-score values. The findings highlight the effectiveness of BPNN in refining predictive models, making it a more reliable tool for credit risk assessment in the Nigerian banking sector.