|
|
Context: The rise of identity theft has become one of the most dangerous growing cybercrimes today, particularly as individuals are now digitally on-boarding; therefore, with minimal information provided for identification/verification purposes, traditional rule-based systems cannot identify many of the sophisticated schemes used today such as Deepfakes, Document Forging, Synthetic Identities etc. Fraud detection has been the focus of much research but there is still a large void in the area of data ingestions, specifically in identifying and alerting Identity Theft prior to an account being created through a Real Time Explainable Solution. Fraud detection is a well-researched topic; however, fraud detection at the time of account creation (during the ingestion of data) remains a largely unexplored area where fraud detection is most important. In addition, current fraud detection systems do not have the capability to use hybrid models that can detect multi-modal, synthetic identities, and deepfakes as well as other cross-channel anomalies. Additionally, most current fraud detection systems do not provide an integrated approach of using both supervised and unsupervised methods for detection or include the ability to provide explanations for the decision-making process of the model to combat modern forms of synthetic and AI-based attacks. We present a Hybrid AI Framework which utilizes Supervised Learning, Unsupervised Anomaly Detection, and Explanatory AI (XAI), to identify Identity Fraud prior to Account Creation. This Framework will combine multiple Data Sources (Documents, Biometric Information, Devices, Structured Attributes) to produce Interpretable Risk Scores, utilizing SHAP Values & Rule Based Explanation, allowing Analysts to Identify Alerts & Resolve Them Efficiently. Our End-To-End Design Offers a Scalable, Compliant Solution to Early-Stage Identity Theft Prevention in Financial Services.
|