Fraud Detection with Graph Features and GNN

Actions Panel

Fraud Detection with Graph Features and GNN

Why Amazon, Tencent, Alibaba and eBay are using Graph for Fraud Detection? - Nikita Iserson

When and where

Date and time

Location

Online

About this event

  • 1 hour
  • Mobile eTicket

Outline:

  • Class Imbalance, Label Scarcity & Fidelity (Fraud cases are rare events)
  • Fraud Camouflage - handle context & feature inconsistency (i.e. fraudsters connecting to regular entities)
  • Investigation and Exploration (visual way to connect the dots for the crime case)
  • Anomaly Detection - handle point, structural and contextual outliers
  • Graph Embeddings - combined with NLP, could be used for scalable fuzzy search and entity resolution
  • Explainability & fairness - adding the context and structure for interpretation, rebalancing the data to remove bias.

Identifying fraudulent behaviors is becoming increasingly more complex as technology advances and fraudsters constantly evolve new ways to exploit people, companies, and institutions. The complexity grows as companies introduce new channels, platforms, and devices for customers to engage with their brand, manage their accounts, and make transactions. Graph neural networks (GNN) are increasingly being used to identify suspicious behavior. GNNs can combine graph structures, such as email accounts, addresses, phone numbers, and purchasing behavior to find meaningful patterns and enhance fraud detection.

I will present short Python demo for Fraud Detection with Graph Features:

  • Feature Engineering (Benford law, Tx velocity, Graph)
  • Exploration Data Analysis (Red Flag Bar Charts)
  • Community Detection (Account - Transaction Graph)
  • Model Training and Feature Importance with Graph Features
  • Explainable AI Modeling for Fraud
  • Model Evaluation and Financial Impact

About the speaker:

Nikita is a Lead Machine Learning Engineer at S&P Global with over 10 years of experience in software engineering, data warehouse development, data analytics, and machine learning. He has built demand forecasting, network analysis, recommender systems, digital twins, and much more covering a wide range of industries, including telecom, retail, and banking.

DataTalks.Club is the place to talk about data. Join our slack community!