At the last Salford Systems CART conference, the results of a comparison a various data mining techniques to predict fraud in a Massachusetts automobile insurance database were presented. One version of the analysis was published in the winter 2009 issue of the Casualty Actuarial Society’s journal Variance. In this update, various data mining techniques and software implementations of the techniques will be applied to a Countrywide automobile insurance database. The data mining techniques tested will include: Logistic Regression, CART, CHAID, Random Forest, Treenet and neural networks. The techniques will be used to predict insurance fraud.

*This presentation was given by Louise Francis at the 2009 Salford Data Mining Conference.