MUMBAI, India, June 26 -- Intellectual Property India has published a patent application (202621053489 A) filed by Moresh Madhukar Mukhedkar; Medhansh Shekhawat; Prisha Kumar; Rohit Jagtap; Vedant Warghade; Anurag Kumar; Dr. Shweta Koparde; and Dr Vivek Patil on April 27, 2026, for Gnn Model For Financial Fraud Detection.
Inventors include Moresh Madhukar Mukhedkar; Medhansh Shekhawat; Prisha Kumar; Rohit Jagtap; Vedant Warghade; Anurag Kumar; Dr. Shweta Koparde; and Dr Vivek Patil.
The application for the patent was published on June 19, 2026, under issue no. 25/2026.
Abstract: The present invention proposes an GNN Model for Fraud Detection System using Machine Learning and Graph Neural Networks, designed to automate and enhance the process of identifying fraudulent financial transactions in digital systems. The system analyzes transaction data such as transaction amount, timestamp, user behavior, location, device information, and relational interactions between entities to determine whether a transaction is fraudulent or legitimate. The system operates through a structured pipeline that includes data preprocessing, feature engineering, anomaly detection, and graph-based modeling. Data preprocessing involves cleaning the dataset, handling missing values, normalizing numerical attributes, and encoding categorical variables to ensure accurate and consistent analysis. Feature engineering techniques are applied to generate meaningful attributes such as transaction frequency, behavioral deviations, and locationbased anomalies, which significantly improve model performance. The processed data is then analyzed using a hybrid approach that combines anomaly detection algorithms and Graph Neural Networks. The anomaly detection component identifies unusual transaction behavior, while the graph-based model captures relationships between entities such as users, accounts, and transactions to detect coordinated and network-based fraud patterns. The system generates output in the form of fraud classification (Fraudulent or Non- Fraudulent) along with a confidence score, providing transparency and support for decision-making. The proposed system improves detection accuracy, reduces false positives, and enhances efficiency compared to traditional fraud detection methods. The invention provides a scalable, intelligent, and data-driven solution that can be integrated into banking systems, payment gateways, and financial platforms for real-time fraud detection. Overall, the system contributes to improved financial security, reduced financial losses, and enhanced trust in digital financial services.
Disclaimer: Curated by HT Syndication.
