Fraud Detection

I did a proof-of-concept in fraud detection once.

Basically, we created a lot of additional attributes from the transaction data (like the number of transactions on one day compared with the number of monthly and weekly transactions, the transaction amount compared to averages, if the amount was a round number, if the same transaction was repeated, if the partner in the transaction was new or existing, if there was a security related event like password change).

The best performing learners were W-RandomForest, AdaBoosted Decision Tree and SVM.

Thanks Balázs Bárány for the reply!

You have a data set and can share to test in RapidMiner?

Regards,
FLávio Loureiro

Sorry, this was a consulting project at a client. I can't share more details.

I came across this article that thoroughly answers what automobile insurance fraud is. Its elements and types? How to fight it? And what challenges in detecting fraud in automobile insurance?

first create basic logistic regression model, later you can change models and optimize parameters and use best out of that.