MUMBAI, India, Aug. 17 -- Intellectual Property India has published a patent application (202621052344 A) filed by Dkte Society'S Textile And Engineering Institute; Ms. Prachi Popat Chougule; Ms. Shital Sanjay Gaikwad; and Ms. Vanchala Bhagwat Sutar on April 24, 2026, for Ai-Driven Fraud Detection System With Temporal Forensics And Adaptive Thresholding.
Inventors include Ms. Prachi Popat Chougule; Ms. Shital Sanjay Gaikwad; and Ms. Vanchala Bhagwat Sutar.
The application for the patent was published on August 14, 2026, under issue no. 33/2026.
Abstract: AI-Driven Fraud Detection System with Temporal Forensics and Adaptive Thresholding This invention describes an AI-driven real-time fraud detection system for financial networks, integrating machine learning, temporal analytics, zero-trust validation, adaptive scoring, and federated learning reconciliation. A Random Forest inference engine classifies transactions based on attributes and contextual metadata, while Layered Temporal Forensics evaluates behavioral patterns across short-, mid- , and long-term intervals to detect anomalies. A Trusted Signal Validator enforces zero-trust checks on device fingerprints, IP entropy, and session continuity, elevating risk upon inconsistencies. Adaptive Threshold Modulation dynamically adjusts fraud-score cutoffs in response to geographic, behavioral, and merchant-related triggers, thereby minimizing false positives and negatives. A Federated Misclassification Reconciliation framework enables privacy-preserving corrections through encrypted feedback, allowing local recalibration without exposing raw data. The system operates through a lightweight API with sub-200 millisecond latency, delivering fraud scores, anomaly reasoning, and evidentiary logs. The architecture provides a transparent, adaptive, and scalable solution for fraud detection in high-volume digital financial ecosystems.
Disclaimer: Curated by HT Syndication.
