Subscribe to DSC Newsletter

Dear all,


I have developed a model using logistic regression (Through probability score) to predict the suspicious claims in the general insurance for further investigation. Now, question is how can I identify the new methods fraud stars may come up with to commit fraud in the industry?

One of the simple ways I am aware of is to refine the model periodically, can you please suggest me any other methods to detect the same?




Shounak Ghosal


Views: 1384

Replies to This Discussion

Dear Kumud,

Thank you for your suggestion. Let me try applying unsupervised techniques in my case. I will keep you updated about my findings here.


Shounak Ghosal
Hi Goshal,

Even me want to know about predicting insurance frauds....can u pls suggest me, how can it to be done by SPSS Modeler...

Thanks you,
Ashok B.
As you noted, the fraudsters are always trying to "reinvent" themselves and avoid detection. Models have and will continue to do a good job at predicting a suspicious claim. But what can an investigator do to analzye this, eliminate false postives and uncover new, hidden methods that are being used to beat the system? Non-obvious Relationship Analysis is a form of data visualization that employs link and node diagrams which can be used to identify hidden patterns in the data that the model may miss. Here's an example of how this works.

Screen one shows a heatmap where the money that is "at risk" is mostly in one business line and attached to one fraud alert -- high appraisal loan. These alerts could have come from the model you built to predict fraud:

Once you identify this, you can analyze the same data in the form of a link analysis diagram. The example below shows the customers linked to property they own and the LOAN OFFICER who granted the loan. Notice the loan officer is in Los Angeles and the proptery is in NY. Odd, isn't it?:

Now that you know this, you may want to investigate other fraud alerts associated with the loan officer's customers. Is there a new pattern of behavior? Do any of the customers have negative scores coming back from real time identity verification services? Who are these customers linked to? Is there a timeline of events that shows the customer depositing larger sums of money into bank accounts?

A good example of this in insurance can be found here:

If you would like to see videos that show how this type of analysis can be done in a matter of minutes, go here:

Good luck.

Tony Agresta

All of these question


On Data Science Central

© 2021   TechTarget, Inc.   Powered by

Badges  |  Report an Issue  |  Privacy Policy  |  Terms of Service