Car Insurance Risk Assessment with Data Mining for an Iranian Leading Insurance Company
Seyed Behnam Khakbaz · International Journal of Business and Economics Research · 2014
Today’s competitive market leads industry to a serious fight. This fight has guided some companies to a sightless selling. Insurance companies lose lots of money each year because of not profitable and risky customers which are attracted blindly. Risky customers are one of the most important treats to insurance companies; therefore some of these companies adopt a credit scoring and risk assessment approach for identifying profitable and risky customers. One of the most preferable methods for risk assessment is data mining. In this article, authors would demonstrate a risk assessment problem in an Iranian leading insurance company. Car insurance customers of this company have been analyzed with six different data mining algorithms (C5, Classification and Regression Tree, Neural Networks, Logistic Regression, Bayesian Networks and Support Vector Machines) in two different approaches. One of these approaches is a direct approach in which the target field (risk) is predicted directly with data mining algorithms and then an ensemble model comprised from them. The other one is an indirect approach in which the target field would be divided in five fields, then five different ensemble models is comprised for each new target field. Afterwards the model with the highest confidence predicts the target fields for a test data record. At the end of this article the better results of indirect model would be shown.