Improving Credit Score Classification Using Long Short-Term Memory and Support Vector Machines Tuned with Whale Optimization Algorithm

Credit scoring is important in the financial industry. While traditional credit scoring has been shown to be useful, machine learning is being explored as an alternative by their ability to find complex patterns that traditional models might miss. This study aims to use the Whale Optimization Algorithm (WOA) for hyperparameter tuning to improve the performance of the Long Short-Term Memory (LSTM) and Support Vector Machine (SVM) model in credit score classification, precisely on the Kaggle Credit Score Classification dataset. Hyperparameter tuning is crucial but can be time-consuming and lead to poor selection. The results show WOA's success in optimizing both models with a performance increase of 1- 4% on each evaluation metric. WOA also surpassed Genetic Algorithm (GA) in tuning LSTM and was comparable to GA in tuning SVM. Despite the study's constraints, WOA found more optimal hyperparameters for each model and outperformed GA, indicating the potential for improved credit scoring.

I. Introduction

Credit score is the main component used in the modern financial system and plays an important role in the decision-making process for granting loans. Credit scoring assists lenders or other financial organizations in determining whether a specific individual or organization can satisfy its credit commitments. This information is important in making decisions about borrowing money or offering credit, as well as for fair treatment of consumers and limiting risks for lenders. It is this precise credit score that can help financial institutions such as banks reduce credit risk and improve their overall financial stability. Currently, supervision of financial regulations regarding lending is increasing, this is also the impact of the global financial crisis that occurred in 2008 [1] which caused large losses, so a credit assessment system is needed that is more reliable and effective in predicting criteria for credit recipients.