Guardian - fraud detection using graph analysis
summary
Guardian ingests company and ownership records into Postgres then builds graphs in Arango or directly in Python, aggregates monthly transaction amounts and invoice counts as weighted edges, and computes centrality and cluster features for models. Orchestration uses Airflow, graph code uses NetworkX, feature selection uses XGBoost, visualization uses Cytoscape. A configurable pattern library flags missing trader and other schemes, sparse labels come from expert-verified pattern matches, and logistic, SVM and random forest prototypes run against those labels to prioritize cases and shorten investigation time.
AI-written summary · about
chapters (9)
- 0:00Guardian fraud detection intro
- 3:36Fraud detection challenges
- 7:29Graph analysis for fraud
- 10:56Building the graph network
- 13:03Suspicious behavior patterns
- 15:14Machine learning for fraud
- 17:46UI and case management
- 20:58Exploring fraud cases
- 22:04Future enhancements
related talks
56mGraph Analytic Systems
26mAdvancing Financial Fraud Detection with Graph Dbs
28mA Quantum Leap In Fraud Detection Accuracy Using Graph-based Features
45mApplications of Graph in Fraud Detection
27mEnhancing Financial Analytics With Event-Driven Architectures and Graph Neural Networks
12mCloud-Native Friendly Fraud Detection
46mGraph Based Detection and Response with Grapl
24mHunting Criminals with Hybrid Analytics