A Fraud Detect System Based on Customized ML Models
Many merchants use user reputation and life time value (LTV) to make transaction, campaign and offer decisions. Although LTV concept can be very useful for pricing and refund policies, we find applying the knowledge to fraud detection might run into problems like over-penalizing new users and not being able to stop account take over (ATO) frauds effectively. Alternatively, we build user profile to provide the easy access to user history for real time decision. Beyond ATO detection, the practice can be applied to determine if a rejected, refunded or chargeback transaction is due to friendly or malicious fraud.