Fraud Detection Graph

Requirements

• Docker
• Python 3
• Git LFS
• Bash

UNIX systems only!

Setup

Clone the repo

git clone https://github.com/sregucki/fraud-graph.git
cd fraud-graph

Install requirements

Run ./setup.sh

Manual setup

Copy CSV files to neo4j/import

sudo cp data/*.csv ./neo4j/import/

Commands for creating the graph

LOAD CSV WITH HEADERS FROM 'file:///accounts.csv' AS row FIELDTERMINATOR ','
CREATE (acc: Account {id: row.id, name: row.name, email: row.email})-[r:LOGGED_ON]->(dev: Device {id: row.device_id})

LOAD CSV WITH HEADERS FROM 'file:///transactions.csv' AS row FIELDTERMINATOR ','
MATCH (source: Account {id: row.source_account_id}),(target: Account {id: row.target_account_id})
CREATE (source)-[r: SENT_TO {id: row.id, amount: toFloat(row.amount), timestamp: datetime(replace(row.timestamp, " ", "T"))}]->(target)

TODO

• Wykrywanie transakcji w pętlach (np. A -> B -> C -> A)
• Kilka kont, jedno urządzenie (done, fix data generation)
• Communities detection (wąska grupa robiąca transakcje jedynie ze sobą)
• Maybe location based detection