Fraud Detection System using Deep Neural Networks

My presentation slide in Re-Work Deep Learning Summit at Singapore, 2017

Transcript

1. Fraud Detec+on System using Deep Neural Networks Hendri Karisma hendri.karisma@gdn-commerce.com / hendri.k.id@ieee.org
2. Fraudulent Transac+on Payment Fraud (phishing, account take-over, carding) System abuse (promo, content, account, logistic and payment methods especially COD) Fraud not only result in financial losses but also produce some reputational risk. Some security measures has been taken by bank or another multinational finance service. [E. Duman et al, 2013]
3. State of the art •  Some methods used in Fraud Detec+on research area: –  GASS 82.78%-91% and MBO algorithm 91.3%-94.35% –  ANN 91.74% –  SVM 83.06% [E. Duman et al, 2013] –  Copula-based method, extreme outlier eliminaUon, PCA, naïve bayes , regression logisUc, k-nearest neighbors, etc.
4. Annual Reports Cybersource
5. Annual Reports Cybersource
6. State of the art BS (bivariate statistics) Feature Extractions PCA (principal component analysis) Information Gain PCA + IG = GPCA Etc.
7. Why Deep Learning •  High nonlinearity Dataset •  Amount of data •  A lot of features •  Mostly unlabeled data
8. Deep neural networks
9. Standard neural networks
10. Standard neural networks
11. Back-propaga+on
12. Deep neural networks [H. Karisma et al,2016]
13. Pre-training •  Denoising auto-encoder •  Restricted Boltzmann machine
14. Auto-encoder
15. Pre-training 1 2 3
16. Deep neural network for repricing gap forcas+ng (bank) •  Equals network topology •  High nonlinearity •  Almost all a_ributes have conUnuous values •  Using auto-encoder •  Minimum mean squared error : 10-9 0.00 0.20 0.40 0.60 0.80 1.00 1 31 61 91 121 151 181 211 241 271 301 MSE (10^-4) Iteration (10^2) SB DNN [H. Karisma et al,2016]
17. Network topologies [H. Karisma et al, 2016]
18. Algorithm Parameters •  Minimum mean squared error : 10^-8 •  Learning rate : 0.75 •  Momentum : 0.5 •  Topology network : equals for each hidden layer •  Hidden Layer : 3 Hidden Layer •  Neuron/Hidden Layer : (26, 26, 26) •  AcUvaUon funcUon : sigmoid funcUon •  Auto-encoder (pre-training) parameters : –  Minimum squared error : 10-5 –  Max epoch: 2000 –  Learning rate : 0.5 –  Momentum : 0.75 –  AcUvaUon funcUon : sigmoid funcUon
19. Dataset •  Dataset : 4000 •  Fraud : 32 (confirm fraud) •  Good transacUon : 2000 •  False address cases : 2157 •  Suspect transacUon : 500 •  A_ributes : +/- 102 •  Non-linearity : High
20. Feature engineering •  Order informaUon (customer info, billing info, shipping info, items, item category, amount, discount, etc) •  Card VerificaUon number (for BIN number) •  Postal address •  Google maps lookup (distance between shipping and billing) •  Telephone area lookup •  Credit history •  Customer order history •  Order velocity monitoring •  IP GeolocaUon •  Value Similarity (shipping and billing address, customer email and customer name)
21. Feature engineering (Velocity) •  Mask card number given : billzip, cus+p, email, name, shipzip. (just count) •  Mask card number given : billzip, cus+p, email, name, shipzip. (changing) •  Email given : billzip, cus+p, name, masked number, shipzip (just count) •  Email given : billzip, cus+p, name, masked number, shipzip (changing) •  So on for billzip, cus+p, name, and shipzip. Then compute the gradient.
22. Feature engineering It will add more than 60 features to dataset. •  Look-up features •  Velocity features •  Other preprocessing 0 2 4 6 8 10 0 1 2 3 4 5 6 7 count email given card changing change card Linear (change card) 0 1 2 3 0 2 4 6 8 count email given chard changing change card Linear (change card) 0 1 2 3 4 5 6 0 5 10 15 20 25 30 count email count of transac+on
23. Result •  Accuracy : 89.475 •  Confusion Matrix •  MSE : 8.31 x 10^-6 Fraud Good predict/actual 1636 364 Fraud 57 1943 Good 0.00 0.20 0.40 0.60 0.80 1.00 1 26 51 76 101 126 151 176 201 226 251 276 MSE (10^-3) ITERATION(10^2) Deep Neural Network for FDS
24. Challenges •  Unbalancing dataset •  Fraud is transacUon perspecUve to fraud is person perspecUve (data structures changing) •  Event data (from checking page, order unUl transacUon/ checkout) •  GPU opUmizaUon •  Network model architecture •  Social network features (text and network) •  Restricted Boltzmann machine or another pre-training •  Graph theory
25. THANK YOU Any question?