MUMBAI, India, Feb. 13 -- Intellectual Property India has published a patent application (202641007460 A) filed by G Ashwin Prabhu; Reeba Rose L; Ricky Jasper K; Subakanth R; Meghana Sai P; Naveensri D; Mohamed Rafiq M; and Karthick Raja D, Chennai, Tamil Nadu, on Jan. 26, for 'a graph based semi - supervised model for flagging suspicious accounts in anti - money landering systems.'
Inventor(s) include Reeba Rose L; Ricky Jasper K; Subakanth R; Meghana Sai P; Naveensri D; Mohamed Rafiq M; and Karthick Raja D.
The application for the patent was published on Feb. 13, under issue no. 07/2026.
According to the abstract released by the Intellectual Property India: "The present invention relates to an Artificial Intelligence (AI)-assisted method and system for detecting and flagging suspicious financial accounts in Anti-Money Laundering (AML) systems, enabling accurate identification, classification, and risk assessment of illicit financial activities from large-scale transactional networks. The system integrates transactional data, Know Your Customer (KYC) information, and domain-specific risk indicators to perform comprehensive analysis of financial behaviors across banking and digital payment ecosystems. The proposed system comprises: (a) a data preprocessing and integration module configured to construct transaction graphs from multi-source financial datasets, where accounts are represented as nodes and monetary transfers as edges; (b) a graph representation learning engine employing semi-supervised Graph Convolutional Network (GCN) variants to generate high-dimensional embeddings that capture relational, multi-hop, and circular transaction patterns; and (c) a hybrid AI-based classification model utilizing XGBoost to combine graph embeddings, tabular behavioral features, and risk profiling scores for suspicious account prediction. The system is trained and evaluated on a combination of real and synthetic AML datasets, including FATF-based typologies, achieving a detection accuracy of up to 92%, an F1-score of 0.87, and a false-positive reduction of approximately 7% compared to traditional rule-based AML systems. Additionally, the model provides interpretable outputs by highlighting transaction paths and risk factors contributing to each alert, improving analyst trust and investigation efficiency. The system further incorporates an adaptive risk profiling mechanism that allows periodic updating of risk indicators in response to evolving laundering typologies. The proposed framework supports scalable deployment through batch or near real-time processing environments, making it suitable for banking institutions, financial intelligence units, and regulatory compliance systems, while aligning with international AML standards and regulatory guidelines."
Disclaimer: Curated by HT Syndication.
