AI-Driven Fraud Detection Using Self-Supervised Deep Learning for Enhanced Customer Identity Modeling

Abstract

The quick growth rate of e-financial services has augmented the need of smart and dynamic fraud detection framework which had the capacity to acquire intelligent lean and dynamic and unbalanced threat models. The given article introduces an AI-based customer identity modeling and fraud system, which presupposes the implementation of hybrid temporal and graph analytics and relies on the self-supervised deep learning to increase the focus on anomalies in the intricate setting of transactions. The suggested system entails contrastive behavioral representation learning and masked feature reconstruction throughout the pretraining stage to acquire discriminative latent embedding of the customer behavior with a minimum requirement of frail information with labeling. These embeddings are trained on a Temporal Transformer that is learned to learn the dynamics of sequential spending and a Graph Neural Network (GNN) that is learned to learn cross-account identity relationships.

It is a fusion of the temporal, identity and self-supervised embeddings, which apply to provide sound decision making by using the stacked ensemble of XGBoost, deep neural networks and logistic regression meta-learner to rank real time risks. The performance metrics on the framework are measured on large and very imbalanced real- world financial transaction data on the accuracy, recall, F1-score, ROC-AUC, the false positive and latency of detection. The recall, precision, and F1-score of the system are 17, 12 and 19 percent higher than the traditional rule- based and supervised fraud detectors, as explained in the experiments. Moreover, identity-conscious modeling will be capable of minimizing the false positive rates by 15 percent and identify it in time, which may be considered in real-time cybersecurity analytics. In general, the proposed framework can be described as a flexible, sustainable and intelligent system of proactive fraud detection, dynamic customer identity profiling, and next-generation financial cybersecurity systems.