accumulo summit 2015: fraud analytics using accumulo, julia and fast sql [leveraging accumulo]
TRANSCRIPT
Copyright © 2014 by Argyle Data Inc. All Rights Reserved. 1
2014 the Year the
Fraud and Cyber-Security
Dam Broke
Rewriting the Book on Fraud
Analytics
Copyright © 2014 by Argyle Data Inc. All Rights Reserved. 2
Stack ShiftHadoop
Accumulo
Presto
Julia
Packet Ingestion
Copyright © 2014 by Argyle Data Inc. All Rights Reserved. 3
Do What Couldn’t Be DonePetabyte Scale
100m Inserts/Second
Interactive SQL Queries
Real-Time Machine Learning
See Everything Non-Intrusively
Copyright © 2014 by Argyle Data Inc. All Rights Reserved. 4
Mobile Fraud $46.3 Billion
Subscription $5.2B
Wangiri $2B
SMS Phishing $1.7B
PBX $4.4B
Roaming $6.1B
Premium Rate 4.7B
IRSF 1.8B
Arbitrage $2.2B
Copyright © 2014 by Argyle Data Inc. All Rights Reserved. 5
Grab from the Network
Feature/Floop Generation
Combine Network and Business Data
Apply the Right ML Algorithms
Copyright © 2014 by Argyle Data Inc. All Rights Reserved. 6
Prove Real-Time Fraud Analytics
at National Scale
World’s Largest Mobile Carrier
2% of Revenue to Fraud
4 Month Rollout
Numerous Multi-Million Attacks
Thwarted
Copyright © 2014 by Argyle Data Inc. All Rights Reserved. 7
Julia
fresh start for scientific compute
speed of C meets / dynamism of Ruby
Copyright © 2014 by Argyle Data Inc. All Rights Reserved. 8
Non-parametric Methods
No / Low Assumptions
but
Computationally Untenable
Copyright © 2014 by Argyle Data Inc. All Rights Reserved. 9
Non-Parametrics in
distributed Julia!
good news
Accumulo doesn’t melt!