Fraud Detection: Credit Card

Business challenge

  • Credit card network’s in-house fraud detection is slow and not accurate enough
  • Model is updated on a yearly basis, and running on a large dataset takes over 2 days


  • Used sophisticated “supervised” and “unsupervised” methods together to augment the existing system for higher accuracy
  • speed allows the model to be updated daily, and the scoring is made in real time

Business benefit

  • 5% increase in fraud detection

Frequent updates allow the capture of changes in fraud patterns; no need to wait for the next yearly update