Fraud Detection Algorithm

[viveksengar]

I wanted to design and code the fraud detection system for credit, debit and ATM Card. I have done some studies and come up with some ideas.

In the different algorithms of AI if i simply use the Bayesian Algorithm or Nural Network approch or even hybrud Approch to calculate the different statical parameters of past transections like IP address, login time, MAC Address, Transection Amount etc.

But even if after traening the system if the Frequency distribution of genuine and fraud cases overlaps a lot. then how can i find the solution to find the fraud in that cases.
Overall i wanted to know how can divide the entire set of data for training to determine the fraud.

Is there any better algorithm or approch..??...Please suggest.