AI-DRIVEN FRAUD DETECTION AND PREVENTION USING HUMAN BEHAVIOR ANALYSIS TO ENHANCE US SOCIAL AND FINANCIAL SECURITY

Abstract

The paper proposes an innovative system comprising of behavioral biometrics and transaction risk modeling to identify fraud on U.S. financial and social sites. The system combines keystroke, mouse/touch gesture, navigation, and social interaction cues into an active behavior integration, which is scored by a mixture-of-experts architecture through drift adaptation. An equation is defined of a patentable Behavioral Trust Vectorization Engine (BTVE), a dynamically risk-posture-weighted embedding dimension. On synthetic financial-social data of the United States, with latency of approximately 8.2 ms, the system can recall 96.5 percent, with precision of 92.0 percent and false positive rate of less than 1.6 percent. An example of an application use-case is a use case that detects a disguised account takeover attempt in real-time. The approach is strong in adversarial mimicry drift.