MUMBAI, India, Aug. 17 -- Intellectual Property India has published a patent application (202641095332 A) filed by Vardhaman College Of Engineering on August 06, 2026, for Decentralized Finance (defi) Fraud Detection Using Graph-Based Machine Learning.
Inventors include Dr. Vasantha S V; Ms. Keerthi Pendam; Ms. Kolarkar Ashlesha Vijay; Ms. Baini Anusha Rani; Mr. Adavelli Ramesh; and Mr. Sankala Sowjanya.
The application for the patent was published on August 14, 2026, under issue no. 33/2026.
Abstract: Decentralized Finance (DeFi) Fraud Detection Using Graph-Based Machine Learning is the proposed invention. The disclosed invention is a decentralised finance (DeFi) fraud detection system with a Temporal Graph Transformer Network (TGTN) for intelligent analysis of blockchain transaction networks. The proposed framework models wallets, intelligent contracts, decentralised exchanges, liquidity pools and digital resources as nodes in a dynamic heterogeneous graph, and blockchain transactions and contract interactions as temporally varying edges. We use transformer based self-attention mechanisms to process real-time blockchain data and generate graph embeddings that capture structural relationships and temporal behavioural patterns. The learned representations allow to detect sophisticated frauds such as flash loan attacks, rug pulls, wash trading, phishing wallets, liquidity manipulation, and coordinated malicious activities at an early stage. An adaptive fraud scoring engine using graph embeddings, behavioural analytics, smart contract vulnerability indicators, and historical transaction profiles to generate explainable risk scoring for wallets and transactions. The framework continuously updates its learning model to adapt to new fraud tactics and offers clear attention-based explanations for each prediction. The invention enhances the security of blockchain, improves the accuracy of fraud detection, reduces false positives and enables secure, scalable and trustworthy decentralised financial ecosystems.
Disclaimer: Curated by HT Syndication.
