Predictive Machine Learning Models for Detecting Financial Fraud and Credit Risk Across Digital Banking Platforms

Abstract

The rapid expansion of digital banking has transformed financial service delivery through mobile applications, online banking, instant payments, digital lending, and API-driven financial ecosystems. However, increased transaction volumes, interconnected platforms, and evolving customer behaviours have intensified exposure to financial fraud and credit risk, challenging conventional rule-based monitoring and static credit-scoring approaches. This study presents a predictive machine learning framework for detecting fraudulent activities and assessing credit risk across digital banking platforms. The framework integrates transactional, behavioural, demographic, account, device, and credit-history data to construct risk-sensitive features capable of identifying complex patterns associated with anomalous transactions and potential borrower default. Supervised and ensemble learning models are applied to fraud classification and credit-risk prediction, while class-imbalance handling, feature selection, hyperparameter optimisation, and cross-validation strengthen predictive robustness. Model effectiveness is evaluated using precision, recall, F1-score, ROC-AUC, false-positive rate, and predictive accuracy. By combining fraud detection and credit-risk analytics within an integrated predictive architecture, the proposed approach supports earlier risk identification, adaptive decision-making, reduced financial losses, and more resilient digital banking operations across increasingly complex financial environments.