Data Skeptic Fraud Detection in Real Time
6 snips
Aug 18, 2020 Fraud detection in e-commerce is a complex challenge, and humor lightens the mood as insights unfold. The importance of converting historical data into actionable datasets for machine learning is explored. Real-time data processing is emphasized as crucial for effective fraud detection. Additionally, the episode dives into managing chargebacks in high-risk industries and the need for agile responses to fraud. Through anecdotes and expert tips, listeners gain valuable knowledge on improving their fraud detection strategies.
AI Snips
Chapters
Transcript
Episode notes
The Seed Scam
- Merchants mail unsolicited seeds from China to generate verified reviews on e-commerce platforms.
- This allows them to boost their products' visibility through seemingly legitimate reviews.
Fraud as Classification
- E-commerce fraud detection is essentially a classification problem, but with delayed and incomplete labels due to chargebacks.
- The nature of chargebacks creates inherent survivor bias in the data.
Look Beyond Transactions
- Don't just rely on historical transaction data for fraud detection.
- Incorporate broader user behavior and metadata like shipping/billing address matching.