THE USE OF ALTERNATIVE DATA IN CREDIT SCORING: PERFORMANCE, FAIRNESS, AND REGULATORY CONSTRAINTS

This paper will study how alternative data can be utilized in credit scoring with respect to its influence on
predictive effectiveness, fairness, and adherence to regulatory limitations. Conventional credit scoring models
usually use only limited financial histories, which are incapable of covering underserved groups and miss subtle
risk indicators. In order to overcome these limitations, this study incorporates other sources of data, such as
utility payments, mobile phone data, and metrics derived by means of social media, into machine learning-based
credit score models. Performance is measured in terms of standard measures that include accuracy, ROC-AUC
and F1 score and fairness is measured in terms of statistical parity and disparate impact measures to reveal the
possible bias according to the demographic groups. Also, the paper takes into account regulatory effects,
including examining how alternative data use fits into frameworks, including GDPR, FCRA, and new fintech
compliance standards. Findings show that alternative data can substantially increase predictive accuracy at
realistically quantifiable fairness trade-offs that need mitigation measures. The results indicate that there should
be a balance between innovation and ethical and legal obligations, and provide practical advice to credit
providers and regulators that want to increase financial inclusion but do not jeopardize compliance. Altogether,
this research will help to get a subtle idea of the way in which alternative data can redefine the process of credit
evaluation in an effective and responsible way.