Enhancing loan accessibility for female entrepreneurs using Generative AI
Open Access DepositedThere have been several studies on the existing gender gap in access to finance. Many female borrowers lack access to credit globally and often receiving less favorable loan outcomes compared to male borrowers. Artificial intelligence (AI) models which are trained on historical lending data containing these disparities risk amplifying the gender bias. This research explored how Conditional Generative Adversarial Networks (CGANs) can be used to mitigate gender bias in predictive credit models. A peer-to-peer lending dataset containing borrower demographic and financial attributes was used. A CGAN framework was developed and conditioned on gender to generate synthetic female entrepreneur records which mirrored the statistical characteristics of the real female entrepreneurs. Baseline machine learning regression models were trained on both imbalanced and balanced datasets and then compared with augmented models retrained using CGAN-generated data. For model evaluation, predictive accuracy, cross-validated generalizability, and fairness metrics (SHAP value differences and error-based metrics), were considered.The statistical test results showed that gender played a role in determining loan amount in the data. After synthetic female borrower profiles were incorporated into baseline models, representation and stability improved. The fairness gap narrowed by at least 23 percent in the augmented models. The gender feature was no longer among the top predictors of loan amounts. Overall accuracy remained similar, however, cross-validated R² improved by more than 15 percent in augmented models, demonstrating stronger generalizability. This work shows that CGAN-based augmentation can function as a practical de-biasing approach, improving fairness and stability in loan modeling without sacrificing predictive performance.
- All rights reserved
Notice to Authors
If you are the author of this work and you have any questions about the information on this page, please use the Contact form to get in touch with us.