Applying Data Mining Technology for Insurance Rate Making: An Example of Automobile Insurance

In this paper we discuss the use of modern data mining (DM) methods to design risk-based insurance premiums for motor vehicles. Our objective is to predict the likelihood and expected value of future claims for each insured based on a myriad of attributes available in the database on the customers and their "peers." The model results may then be used for underwriting and for rate making. We employ a two-stage approach, involving a survival analysis model and a linear regression model, to estimate the risk level of each customer and the proneness to file a claim. The study was performed on actual data set obtained from a small insurance company. We demonstrate our ability to discover new underwriting parameters, build accurate predictive models and to distinguish between distinct groups of policies. The new method creates a new ordering of the policies where the most risky people were, on the average, 12 times more expensive than the least risky people. The importance of the study is not in the particular results, which are specific for the particular company and its environment, but rather in the demonstration of the general ability to use data mining for insurance rate making purposes, and in the original use of the concept of survival analysis and the concept of mean time between claims for this purpose.