Real-time fraud detection using delta live tables and machine learning
Year: 2025
Type: article
Abstract: This article examines the transformative impact of Delta Live Tables (DLT) integrated with machine learning techniques on real-time fraud detection in financial institutions. Traditional batch processing approaches create critical vulnerabilities through delayed detection, while DLT offers a declarative framework that drastically reduces processing latency and improves detection accuracy. The article analyzes multiple dimensions of this technological shift, including architectural design, machine learning model performance, implementation strategies, and business benefits. Various machine learning approaches—from anomaly detection techniques like Isolation Forest and autoencoders to classification models such as Random Forest and neural networks—create a multi-layered defense system when deployed within DLT pipelines. The article outlines a comprehensive implementation architecture comprising data ingestion, feature engineering, scoring, decision-making, and feedback loop components. While highlighting significant business advantages including reduced fraud losses, decreased false positives, operational efficiency, improved regulatory compliance, and enhanced adaptability, the article also addresses implementation challenges related to model drift, feature latency, explainability requirements, and processing trade-offs.
Author: Kothinti, Kedarnath Goud
Institution: Liverpool John Moores University
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7107948878
FWCI:
Cites:
Cited by:
Related to: