Fraud Detection Cookbook, PRODYNA
0 likes365 views
This document discusses using graph databases for fraud detection. It provides examples of how a graph database could model complex transactional data from the retail sector to detect fraudulent patterns. The document outlines a three-level approach to fraud detection, including manual rules, automated rule-based detection, and machine learning. It also demonstrates how the graph database allows for relationship exploration, routing, and strategic analysis of connected data.
Fraud Detection Cookbook, PRODYNA
- 1. © PRODYNA 2017 Slide 1 PRODYNA GRAPH DATABASE USE CASES FRAUD DETECTION COOKBOOK
- 2. © PRODYNA 2017 Slide 2 EXAMPLE: RETAIL SECTOR BUSINESS GENERATES COMPLEX DATA
- 3. © PRODYNA 2017 Slide 3 DIGITAL DISRUPTORS 7000 jobs per day 1,5 TB data per day 60 million active users 75+ Countries 500+ Cities 1000s of ops / city Daily / Weekly statistics Real time analythics Up to 24 GB per second 50 million subscribers
- 4. © PRODYNA 2017 Slide 4 DECISIONS INSIGHTS TO REAL TIME DECISIONS § Cross Selling § Dynamic Pricing § Shopping Offers § Portfolio management HOW DATA DRIVES BUSINESS | REAL TIME RESULTS / BUSINESS / WHAT ? SMART SOFTWARE DATA MANAGE LARGE DATA VOLUMES § High performance § Scalable § Real Time INSIGHTS DATA TO INSIGHTS AND PREDICTIONS § Mesure buying patterns § Know your customers SMART ENTERPRISE / MOBILE APPLICATIONS
- 5. © PRODYNA 2017 Slide 5 CORNERSTONES OF SUCESS DIGITAL DISRUPTORS DISTRIBUTED DATA MAP AND REDUCE CLOUD TECHNOLOGIES
- 6. © PRODYNA 2017 Slide 6 THE PROBLEM § Detect fraud § Example: Airline business § Frauds can be simple § Frauds can be very complex § False positives § Short-term bookings for expensive flights § Bookings for other code share partners § e.g. Book via Lufthansa for United Airlines § Booking with unknown credit card in an away-region (PSCC)
- 7. © PRODYNA 2017 Slide 7 STUPID LONG LISTS OF DATA THE SOURCE DATA
- 8. © PRODYNA 2017 Slide 8 IN ORDER TO GAIN SOME INSIGHT WHAT DO WE NEED § Facts § All we can get § Connected! Ø Fortunately Neo4j offers a great way for doing this
- 9. © PRODYNA 2017 Slide 9 § Stupid long lists of data § Facts with relationships § Booking, passengers, flights, regions, time, historic data DATA HAS MORE THAN VALUES COLLECTING ALL FACTS
- 10. © PRODYNA 2017 Slide 10 DIFFERENT SOURCES INTO ONE GRAPH
- 11. © PRODYNA 2017 Slide 11 SEARCHING FOR DATA
- 12. © PRODYNA 2017 Slide 12 WALKING THE GRAPH
- 13. © PRODYNA 2017 Slide 13 …AND FINDING ALL RELATED INFORMATION
- 14. © PRODYNA 2017 Slide 14 RDBMS Neo4j Mathematical model Set theory Graph theory Basic concepts Table Node Relationship Implicit (keys) Explicit (type) Query language SQL Cypher Transactionality yes yes Network access Proprietary REST + BOLT COMPARISON RDBMS AND NEO4J
- 15. © PRODYNA 2017 Slide 15 THE BASIC DATA FRAUD DETECTION
- 16. © PRODYNA 2017 Slide 16 Stolen user account Stolen credit card NOW EASY TO SEE HERE ARE SOME FRAUDS
- 17. © PRODYNA 2017 Slide 17 § Human written rules detect fraud § Rules are improved over time § Deep learning § Human teaches neuronal network § Frauds are detected automatically § False positive are eliminated over time § Graphical interactive viewer MANUAL LEVEL 1 AUTOMATED RULES LEVEL 2 MACHINE LEARNING LEVEL 3
- 18. © PRODYNA 2017 Slide 18 LEVEL 2: RULE BASED FRAUD DETECTION
- 19. © PRODYNA 2017 Slide 19 HOW DO I FLY FROM VIENNA TO LOS ANGELES? ROUTING § Neo4j can find pathes § Very fast § Routing logic
- 20. © PRODYNA 2017 Slide 20 COMPLEX DATA STRATEGIC DATA
- 21. © PRODYNA 2017 Slide 21 WHAT HAS APPLE TO DO WITH GERMANY? STRATEGY
- 22. © PRODYNA 2017 Slide 22 THANK YOU ANY QUESTIONS? VISIT US AT OUR PRODYNA BOOTH
- 23. © PRODYNA 2017 Slide 23 TWITTER @dkrizic
- 24. © PRODYNA 2017 Slide 24 FOR MORE INFORMATION VISIT PRODYNA.COM CONTACT US +49 69 597 724 0 info@prodyna.com Facebook/prodynaag Twitter/prodynaag
- 25. © PRODYNA 2017 Slide 25 DARKO KRIŽIĆ – PRODYNA FRANKFURT PRODYNA IN YOUR REGION § PRODYNA AG § Ludwig-Erhard-Straße 12-14 65760 Eschborn § T +49 69 597 724 - 175 F +49 69 597 724 - 700 M +49 176 178 70 175 § darko.krizic@prodyna.com prodyna.com