How do I determine syndicate and collusion indication with clustering and network analysis on a large unlabeled user transaction data? So far, I've only been training with labeled data on fraud-related users to cluster whether or not another user is suspicious (similar behaviors to fraud-related users) using Random Forest. Or using network analysis (NetworkX on Python) for requested or reported users to see users related to these suspicious/fraudelent users.

How do I implement this on a larger scale (with hundreds of millions of merchants and users) with unlabeled data? Is this possible? Would that confound the betweenness centrality index?

I tried using K-means clustering and get the 3 clusters as its optimal, but I can't seem to classify which of them are the low, medium, or high risk.

Any reference or advice would be helpful. Thank you!



Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.