Real-time Fraud Detection: Challenges and Solutions
summary
Coverage of a real-time fraud pipeline using Hazelcast: ingest from Kafka, cache operational data, run lightweight ML and SQL-based stream processing, and output to alerts or Kafka. Concrete claims: handling 20,000 TPS, linear scaling to 1 billion transactions under 30 ms on 45 nodes, and a 4×–20× speedup in some migrations. Demos show an airport-coordinate time-window rule and a Jet pipeline that builds per-user trend maps and composite scores, autoscaling nodes, three lines to define a pipeline, and open-source deployment.
AI-written summary · about
chapters (18)
- 0:00Introduction to real-time fraud
- 1:10Defining real-time processing
- 2:59Fraud detection in financial systems
- 5:06High-performance payment processing
- 7:13Sub-millisecond fraud detection window
- 9:46Fraud detection architecture overview
- 12:11Hazelcast platform for fraud
- 16:51Scaling fraud detection solutions
- 20:19Giga-scale transaction performance
- 22:08Real-world fraud detection use case
- 23:16Demo: Travel-based fraud detection
- 31:45Demo: Account scoring and trends
- 39:54Q&A session begins
- 41:18Q&A: Sleep function in algorithm
- 44:38Q&A: Time series predictions
- 47:06Q&A: Hazelcast open-source licensing
- 48:37Q&A: Security of open-source tools
- 50:07Closing remarks and future trends
related talks
47mReal-time Fraud Detection Challenges and Solutions
25mReal-Time Fraud Detection at Scale
13mBuilding Financial Fortresses: Real-Time Fraud Detection
27mBeyond Logs: Real Time Anomaly Detection with Machine Learning
10mReal-Time Fraud Prevention at Scale
13mTransforming Financial Services with Real-Time Analytics
13mBuild Real-Time Analytics Platforms for Financial Crime Detection
33mReal-time Trade Data Processing and Machine Learning at Giga-Sca...