AN INTELLIGENT AND SCALABLE DEEP LEARNING ARCHITECTURE FOR REAL-TIME FINANCIAL FRAUD PREVENTION

Read full text Summarise with VIRAT
Authors: Leadership and Sciences International Journal of Technology
Cited by: 0

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Instantaneous detection of financial misbehavior in environments with a large number of transactions made possible by an extensible deep learning architecture. The suggested approach uses streaming data pipelines and hybrid neural networks with recurrent and convolutional layers to detect spatial and temporal fraud tendencies. Horizontal scalability and low-latency inference are guaranteed by a distributed microservices-based strategy, even when dealing with a huge volume of transactions. A less reliance on human rule engineering is achieved by the system through the explicit learning of feature representations from raw transactional and behavioral data. To make it easier to spot rare cases of fraud, an adaptive attention mechanism gives higher priority to signals that pose a higher risk. The model can adapt to new fraud schemes and prevent concept drift because to its capacity to learn online. The administration of uneven and noisy datasets is made easier with anomaly filtering and strong preparation. We use attention visualization and post-hoc interpretability approaches to make models easier to understand. Extensive research on benchmark and real-world datasets has shown that this approach outperforms standard machine learning algorithms in terms of precision, recall, and F1-score. Thanks to its millisecond reaction times and fault-tolerant design, the system is perfect for real-time authorization tasks and can keep running smoothly across different financial infrastructures.

Fields of Study & Subjects

Imbalanced Data Classification Techniques Anomaly Detection Techniques and Applications Privacy-Preserving Technologies in Data

Publication details

Journal / source
IJTLS
Publication date
26 March 2026
Language
EN