Anti-Money Laundering Tracker Using Graph-AI

Money laundering schemes can involve transfers through several connected accounts, so suspicious activity may be difficult to recognize by examining each transaction separately. This paper presents an Anti-Money Laundering (AML) Tracker that combines graph-pattern analysis with graph representation and machine-learning methods to support the examination of potentially suspicious transaction networks. Accounts are represented as nodes and transfers as directed relationships in Neo4j. The project uses the IBM AMLSim synthetic dataset, with approximately 9,945 accounts and 68,289 transactions reported in the project materials. Following data validation and preprocessing, account and transaction records are loaded into the graph database. Cypher queries are used to explore circular transfers, layering-like chains, fan-in and fan-out structures, and multi-hop paths. The described graph-AI pipeline includes FastRP embeddings, Random Forest classification, and GraphSAGE-based neighbourhood learning. Graph visualization helps users inspect connected accounts and candidate suspicious paths. The functional checks documented for the project show that selected graph operations and pattern queries produced their expected outputs. These checks do not demonstrate predictive accuracy; no verified numerical model-evaluation results are available for reporting. The system is intended to assist investigators, not to determine criminal activity automatically. Future evaluation should document model settings, use held-out data, measure false positives, and assess whether findings from synthetic data generalize to real financial transactions.