ARTIFICIAL INTELLIGENCE IN BANKING: CREDIT SCORING AND FRAUD DETECTION

Abstract

The demand for rapid banking decisions is making it necessary to connect credit-risk assessment and fraud detection within one technological system. The practical problem involves inaccurate lending decisions, delayed alerts, and excessive processing time for customer transactions. This study aims to quantitatively evaluate how an artificial-intelligence approach can coordinate credit scoring with fraud monitoring. A quantitative comparative study was conducted using historical banking data; the protocol combined a longitudinal panel study, machine-learning model validation and evaluation, feature engineering and selection, and gradient boosting. Across 225 observations, the credit-risk score had a mean of M=0.47 and SD=0.08, with values ranging from 0.08–0.91. The mean fraud probability was M=0.06, with a range of 0.00–0.88. Mean transaction processing time was M=18.70 milliseconds, while the observed range was 7.20–41.50 milliseconds. The evidence indicates that credit risk and suspicious activity can be monitored within a unified analytical workflow. The study’s scientific contribution is a validated integrated evaluation approach that links two banking functions through a common gradient-boosting model and a shared assessment protocol. The findings support coordinated use of predictive risk indicators rather than isolated decision pipelines. Before institutional deployment, practitioners should test model explainability, sensitivity to data drift, fairness across customer segments, and cybersecurity safeguards. These controls are essential because faster automated decisions may otherwise amplify undetected bias or operational vulnerabilities. The proposed framework therefore provides a measurable basis for designing accountable banking intelligence systems while preserving separate validation requirements for lending and anti-fraud decisions.