Learning The Hows and Whys of Machine Learning
summary
Tebex uses a data set of 18 million payments and iterates: a naive Bayes prototype fails on highly imbalanced data (2.8M good vs ~24k chargebacks), so experiments add resampling, tokenized binary features and move to naive Bayes then k‑NN via the PHP‑ML library. Per‑store models hit ~70% accuracy; per‑customer anomaly detection flags legitimate payments at ~85% (15% false positives) while catching ~31% of chargebacks. The team builds a prototype long‑running decision API and plans weighting, outlier removal and distance‑weighted k‑NN.
AI-written summary · about
chapters (6)
- 0:00Start
- 7:15The Challenge
- 8:30Supervised Learning
- 9:55Classification
- 36:28Non-Sequential Data
- 49:04Per-Customer Detection
related talks
13mHarnessing Machine Learning to Combat E-commerce Fraud
37mThe Role of Machine Learning in Today’s Payment Ecosystems
30mMachine Learning Techniques for Class Imbalances & Adversaries
47mTrain, Evaluate, Repeat: Building a Credit Card Fraud Detection System
39mTrain, Evaluate, Repeat: Building a Credit Card Fraud Detection System
30mMachine Learning: Beyond the Tutorial
16mLeveraging ML to Create a Reliable Anti-Fraud System
45mAnti-Fraud in Reverse and the Use of ML Methods in It