A Machine Learning Framework Integrating Digital Footprints to Expand Credit Access for Credit-Invisible U.S. Populations

• Abbas Ahsun

  Texas University

  Author

Approximately 45 million American adults remain credit-invisible or have thin credit files, systematically excluded from mainstream financial services due to reliance on traditional credit scoring models that fail to capture financial behavior outside formal banking channels. This research addresses the critical gap in fair, transparent, and accurate credit assessment for underserved populations by developing and evaluating a machine learning framework that integrates digital footprint data with fairness-aware and explainable AI techniques. The study employs a design-based research methodology, utilizing publicly available consumer data and simulated alternative credit indicators to train and validate multiple model architectures including logistic regression, random forest, gradient boosting (XGBoost), and neural networks. The proposed framework achieves a predictive accuracy of 89.4% (AUC-ROC = 0.92), reducing disparate impact by 42% compared to baseline models through in-processing fairness constraints. SHAP (SHapley Additive exPlanations) analysis identifies transaction consistency, digital payment history, and utility payment regularity as the most influential predictors of creditworthiness for previously unscored populations. The findings demonstrate that algorithmic fairness interventions can substantially reduce approval disparities across protected demographic groups while maintaining strong predictive utility. This research provides a reproducible, ethically-grounded framework for financial institutions and policymakers seeking to responsibly expand credit access to credit-invisible populations through alternative data and machine learning.

• pdf

Copyright (c) 2026 Abbas Ahsun (Author)

This work is licensed under a Creative Commons Attribution 4.0 International License.