AI-Powered Fraud Detection in Banking: A Big Data Analytics Perspective
TL;DR
Analysis indicates that ensemble models incorporating XGBoost with LSTM networks achieve accuracy exceeding 98% with substantially reduced false positive rates, and critical challenges including model interpretability, data privacy, and deployment scalability are examined.
Abstract
Banking fraud presents a persistent challenge in the digital era, with financial losses continuing to escalate as transaction volumes grow exponentially. This paper examines AI-enabled fraud detection frameworks that leverage big data analytics, machine learning, and deep learning methodologies. The review systematically categorizes existing approaches across supervised learning, anomaly detection, ensemble methods, and deep neural network architectures. Special attention is directed toward hybrid frameworks that combine multiple techniques to address extreme class imbalance, concept drift, and real-time processing constraints. Analysis indicates that ensemble models incorporating XGBoost with LSTM networks achieve accuracy exceeding 98% with substantially reduced false positive rates. Critical challenges including model interpretability, data privacy, and deployment scalability are examined, alongside promising directions for future research in this domain.