A Hybrid Cloud-Integrated Autoencoder-GNN Architecture for Adaptive, High-Dimensional Anomaly Detection in US Financial Services Compliance Monitoring

Abstract

A group that deals with money in the United States must follow very strict rules that are put in place by an agency like the SEC, FINRA, and OCC. To do this, compliance monitoring systems are deployed, especially for discovering activities that are out of the ordinary such as money laundering, insider trading, and fraudulent activities. Conventional rule-based systems impose great difficulties in the sector because of the high number of false positives, the limited scale of operation and the lack of knowledge of the context. These systems often produce excessive false positives; hence compliance digital units get overloaded, and they cannot introduce dynamic management to keep up with the increasing complexities in the financial network and the data volume. This paper introduces a hybrid cloud-based deep learning approach that is designed to combine auto encoders for anomaly detection in transactions and graph neural networks (GNNs) for anomaly analysis on the network. To make the most use of the cloud infrastructure, the solution suggested should not only improve performance but also render real-time anomaly detection possible and boost operational efficiency.