Unsupervised shallow learning for fraud detection on marketplaces
summary
Claim: shallow unsupervised models flag local anomalies in marketplace seller behavior and scale to real traffic. Built a production score pipeline at Adyen using Python, Airflow, Spark and Spark ML random forest to predict past shopper ratio as a proxy label, later experimenting with autoencoder bottleneck neural nets in Keras TensorFlow and ensembles to reduce gradient volatility. Trained on 800 million rows (14 features) in ~36 minutes on 16 executors. Validation uses a crafted test chamber and ongoing migration to graph models with PyTorch Geometric.
AI-written summary · about
chapters (8)
- 0:00Start
- 1:00Machine Learning for Fraud Detection
- 4:15Significant Shopper
- 6:25Clustering
- 15:06Scalability Problems
- 15:37How Do You Validate Unsupervised Learning How Do You Hypertune a Supervised Unsupervised Learning Algorithm
- 16:54Account Takeover
- 23:59How Did You Handle Pi Information like the Kyc Details for the Customers Especially When Storing It in the Graph Database
related talks
18mReal-time Fraud Detection using Deep Learning
16mLeveraging ML to Create a Reliable Anti-Fraud System
52mModern Fraud Prevention Using Deep Learning
13mHarnessing Machine Learning to Combat E-commerce Fraud
22mFighting against Marketplace Scams using Community-driven AI
40mFraud detection without labeled data
37mThe Role of Machine Learning in Today’s Payment Ecosystems
24mMulti-tasking fraud detection From Trees to MLPs