Case Study: French Motor Third-Party Liability Claims

Abstract

We provide a tutorial that compares a classical generalized linear model for claims frequency modeling to regression tree, boosting machine and neural network approaches. We explore these methods, discuss their calibration and study their predictive power on an explicit motor third-party liability insurance data set. The results of the case study show that a simple generalized linear model does not capture interactions of feature components appropriately, whereas the other methods are able to address these interactions.

Keywords: data science, machine learning, predictive modeling, claims frequency, motor insurance, regression trees, boosting machine, neural network, generalized linear models, feature engineering, covariate selection

JEL Classification: G22