Evaluation of Financial Credit Risk Management Models Based on Gradient Descent and Meta-Heuristic Algorithms

Abstract

An efficient credit risk management model is a promising technique that provides Financial Institutions or Banks the ability to determine a creditworthy customer from a non-worthy customer. The fact remains that no country’s economy can survive or improve without credit using historically available data. This paper presents an evaluation of several gradient descent techniques, and metaheuristic optimization algorithms implemented in Machine Learning and Multi-layer perceptron for better credit risk prediction. It also handles imbalanced dataset using smote Edited Nearest Neighbour. The study provided various architectures and advantages of the algorithms while addressing how the limitations can be improved to build a better credit risk model and improve model accuracy. The study showed MLP WOA achieved accuracy of 98.56% based on Adam gradient descent to achieve faster convergence and exploration compared to MLP PSO with 98.39%.