How shubhamdalvi1999/neo4j-graph-fraud-detection works
Project type and purpose
• This is a graph-analytics fraud-detection pipeline built around a single Neo4j database.
• It ingests a Paysim dataset dump, uses Cypher scripts and the Neo4j GDS/APOC plugins to detect first-party (synthetic identity) and second-party (money mule) fraud, and exposes results via Python and Neo4j Bloom for exploration.
File structure
• Top‐level:
• docker-compose.yml – spins up Neo4j with GDS & APOC
• data/ – pre-processed .dump file(s) for import
• code/cypher/ – Cypher modules (data loading, graph projection, WCC, PageRank, node similarity, scoring)
• code/python/ – Python example connecting via Bolt driver
• fraud demo/ – ad-hoc Cypher scripts for generating and visualizing demo scenarios
• Runbooks and README provide operational guidance.
• Key patterns:
• A single graph database holds all nodes & relationships.
• Analytic steps are implemented as discrete Cypher modules (module1…module4).
• External components (Python, Bloom) query the post-processed graph.
README for additional insights
• Confirms core components: Neo4j DB, GDS plugin, APOC plugin, data dump, Cypher scripts, Python driver, optional Neo4j Bloom.
Detail and capture essential elements
• Break the “processing layer” into four sub-steps (module1…4) to reflect the code structure.
• GDS & APOC plugins as auxiliaries attached to Neo4j.