cancel
Showing results for 
Search instead for 
Did you mean: 

Use Case: Fraudulent claim detection

Use Case: Fraudulent claim detection

Predicting fraudulent auto insurance claim and avoiding losses to improve profitability and retain good customers


What’s the problem?

In 2016, the US Property and Casualty market lost $80+ billion in fraudulent claims. These false claims adversely and directly affect insurance companies in two ways. First, the company loses money by paying for an accident that never happened. Second, the insurers then offload at least a portion of that cost to their good customers in the form of an increased premium. These increasing costs may, in turn, encourage subscribers to seek insurance elsewhere, furthering the company’s loss.

The challenge and solution

Fraudsters are not discriminating—they target any insurance company, large or small, that doesn’t have a predictive system to catch fake claims. Without a predictive model to identify fraud, insurers have two options: 1. Review no claims and simply pay out on fraud claims or 2. Review every claim and lose good customers at renewal time due to poor treatment and slow claims processing. A mid-size insurance company may receive hundreds of claims each day; large insurers may receive hundreds of thousands. Both need a highly accurate model capable of separating fraudulent from non-fraudulent claims, sending questionable claims for investigation, and quickly paying those that are legitimate. With DataRobot’s automated machine learning techniques, an insurer can build highly accurate predictive models based on scoring thresholds and internal business rules. Additionally, the platform can provide “reasons” behind each prediction to aid the claims adjustment team and Special Investigation Unit (SIU) in their investigations.

For example…

Jill is the predictive modeling and actuarial team lead with the Newfoundland Insurance Company. Jill’s long-time colleague in the Claims Management team, Mary, just heard from the CFO—again–about his increasing concern over fraudulent auto claims. Mary, as Head of Claims, is not surprised and thinks that it is the right time for Newfoundland to build a predictive model that will catch fraudulent claims early in the process. Over lunch, the team heads discuss how they should go about building an auto insurance fraud predictive model to successfully identify fraudulent claims and avoid losses. Jill tasks her team with developing these models in DataRobot and, to avoid issues with operationalizing the models later on, involves Mary’s business team in the process.


Training data: https://s3.amazonaws.com/datarobot-use-case-datasets/DR_Demo_Car_Insurance_Fraud_train.csv

Prediction data: https://s3.amazonaws.com/datarobot-use-case-datasets/DR_Demo_Car_Insurance_Fraud_pred.csv

Data Dictionary: https://s3.amazonaws.com/datarobot-use-case-datasets/Insurance+Fraud+Data+Dictionary.pdf

(Originally posted May 2018.)

Version history
Revision #:
6 of 6
Last update:
Friday
Updated by: