Auto Insurance Fraud Model - Derived Variables

dincoo — Tue, 22 Mar 2016 09:00:58 GMT

Hi all,

We are working on a fraud detection project for auto insurance. We have prepared out analytical base table and now we are working on generating derived variables from existing ones.

Do you have any suggestions to us about derived variables?

For example:

Estimated_claim_value / total_payments_amount

Many thanks,

Onur

Re: Auto Insurance Fraud Model - Derived Variables

WendyCzika — Wed, 23 Mar 2016 13:53:13 GMT

Here was a response from SAS employee John Stultz:

In case you have not seen this, here is an old but good paper by SAS’ Terry Woodfield (Predictive Modeling in the Insurance Industry Using SAS Software-- http://www2.sas.com/proceedings/sugi26/p013-26.pdf ) that might help give you some ideas on how to create derived fields and use them as model inputs within an Enterprise Miner process flow.

You might also find more current information/examples in Global Forum papers by searching the Online Proceedings: http://supportprod.unx.sas.com/events/sasglobalforum/previous/online.html

And as always, you can usually find a bunch of stuff by searching the internet. For example, here is a list of some derived/binary variables that might be relevant: (https://www.researchgate.net/publication/227540405_Detection_of_Automobile_Insurance_Fraud_With_Discrete_Choice_Models_and_Misclassified_Claims)

Characteristics of the Insured/Claimant/Policy:

AGE: Age of insured driver when the accident occurred
LICENSE: Number of years since the insured obtained first driver’s license
RECORDS: Number of previous claims of the insured
COVERAGE: Third-party liability equals 1; extended coverage equals 0
DEDUCTIBLE: Existence of a deductible equals 1; otherwise equals 0
ACCESSORI: Coverage for accessories equals 1; otherwise equals 0

Characteristics of the Vehicle:

VEHUSE: Vehicle for private use equals 1; other uses equal 0
VEHAGE: Age of the vehicle

Characteristics of the Accident:

FAULT: Insured accepts the blame for the accident equals 1; other-wise equals 0
NONURBAN: Accident occurred in a nonurban area equals 1; otherwise equals 0
NIGHT: Accident occurred at night equals 1; otherwise equals 0
WEEKEND: Accident occurred during a weekend equals 1; otherwise equals 0
WITNESS: Existence of witnesses equals 1; otherwise equals 0
POLICE: Existence of police report equals 1; otherwise equals 0
ZONE1: Zone with high level of accidents equals 1; otherwise equals 0
ZONE3: Zone with low level of accidents equals 1; otherwise equals 0
REPORT: Existence of a suspicious textual report equals 1; otherwise equals 0. This variable indicates that the claimant reported unusual circumstances for the accident.
NAMES: Same family name for insured and the other vehicle driver equals 1; otherwise equals 0.
PROXIM: Accident occurred between the policy issue date and the policy effective starting date equals 1; otherwise equals 0.
DELAY: Claim not reported to the company within the established period equals 1; otherwise equals 0

topic Auto Insurance Fraud Model - Derived Variables in SAS Data Science

Auto Insurance Fraud Model - Derived Variables

Re: Auto Insurance Fraud Model - Derived Variables