Modern Fraud Prevention Using Deep Learning
summary
Claims practical pipelines that use deep learning plus graphs to find fraud: a graph DB flags rings of shared IDs and addresses, a voice-fingerprint pipeline preprocesses audio into spectrograms and trains a neural model that hit ~97–98% on a tiny dataset, and supervised models (random forest on ~20k Freddie/Fannie records) show credit score explains ~25% while local HPI is most predictive; UK context cites 0.83% mortgage fraud rate and an autoencoder finds outliers for analyst review, but work is limited by noisy labels and dataset size.
AI-written summary · about
chapters (12)
- 0:00Introduction to fraud
- 2:01The cost of financial crime
- 5:59Legal requirements
- 8:31Current detection methods
- 12:19Machine learning basics
- 16:53Types of ML learning
- 22:30Deep learning concepts
- 28:44Digit recognition example
- 37:17Graph databases for fraud
- 41:04Voice authentication demo
- 45:38Mortgage default prediction
- 48:52Anomaly detection
related talks
18mReal-time Fraud Detection using Deep Learning
16mLeveraging ML to Create a Reliable Anti-Fraud System
26mAI-Powered Fraud Detection in Cloud-Native Digital Payments
29mDeep Learning Applications to Online Payment Fraud Detection
08mAI-Powered Fraud Detection in Financial System
12mCloud-Native Friendly Fraud Detection
40mFraud detection without labeled data
26mAdvancing Financial Fraud Detection with Graph Dbs