Anti-Money Laundering Tracker Using Graph-AI (International Journal Of Scientific Research In Engineering & Technology)

Summary

• The Anti-Money Laundering Tracker models transaction networks using the IBM AMLSim synthetic dataset, which includes approximately 9,945 account records and 68,289 transactions.p.1
• Accounts are structured as nodes and transfers as directed relationships in Neo4j to trace multi-step financial flows.p.2
• Cypher queries are executed to detect graph patterns such as circular transfers, layering-like chains, fan-in, fan-out, and multi-hop paths.p.2
• The graph-AI architecture incorporates FastRP embeddings, Random Forest classification, and GraphSAGE neighbourhood learning.p.2
• Functional validation checks passed across graph creation, relationship mapping, and pattern detection modules.p.3

Key insights

1. Reviewing financial transactions as network graphs reveals multi-hop paths and structural flows that individual transaction reviews miss.p.1
2. Graph pattern queries like fan-in and fan-out are easy to inspect but can generate false alerts due to similar legitimate commercial activity.p.2
3. FastRP generates structural node embeddings while GraphSAGE aggregates features from neighbouring nodes in the graph network.p.2
4. Passing functional execution checks verifies software pipeline operations but does not establish predictive model accuracy.p.3
5. Findings derived from synthetic datasets like AMLSim cannot be directly generalized to live banking transaction environments.p.3