Authors:
Rejwan Bin Sulaiman, Sahiti Puppala

Addresses:
Department of Computer and Information Sciences, Northumbria University, London, England, United Kingdom.

Abstract:

The paper has never been more important to accurately forecast borrower behaviour, where a lending decision could risk financial stability. Although the potential of credit assessment models is being explored, they are confined to traditional financial information or a narrower territorial scope. By using data samples from two different lending platforms, where borrowers’ credit histories are frequently unclear and have a broad geographic distribution, this study offers a more comprehensive approach to credit risk analysis. For the investigation, five ML models, namely Logistic Regression, Decision Tree, Random Forest, Gradient Boosting and XGBoost, are employed. The models are trained and tested using a combination of sourced samples and evaluated using metrics such as accuracy, precision, recall, F1 score, confusion matrix and ROC AUC. The findings indicate that ensemble models, particularly XGBoost, are most effective, with consistent performance across all metrics. It is also identified that characteristics such as loan amount, interest rate, instalment amount, term period and demographics such as annual income, gender, home ownership and country of origin influence repayment behaviour. These insights contribute to the field of credit risk assessment with practical implications of reducing default rates that improve lenders’ profitability and stability.

Keywords: Decision Tree; Repayment Behaviour; Lending Platform; Financial Stability; Gradient Boosting; Random Forest; Logistic Regression; Credit Risk Assessment; Instalment Amount.

Received on: 04/06/2025, Revised on: 27/07/2025, Accepted on: 12/10/2025, Published on: 07/03/2026

DOI: 10.69888/FTSFDS.2026.000622

FMDB Transactions on Sustainable Finance and Data Science, 2026 Vol. 1 No. 1, Pages: 19–38

• Views : 56
• Downloads : 12

Download PDF