Modelling claim frequency and claim severity are topics of great interest in property-casualty insurance for supporting underwriting, ratemaking, and reserving actuarial decisions. This paper investigates the predictive performance of Gradient Boosting with Decision Trees as base learners to model the claim frequency in motor insurance using a private cross-country large insurance dataset. The Gradient Boosting algorithm combines many weak base learners to tackle conceptual uncertainty in empirical research. The findings show that the Gradient Boosting model is superior to the standard Generalised Linear Model in the sense that it provides closer predictions in the claim frequency model. The finding also shows that Gradient Boosting can capture the nonlinear relation between the claim counts and feature variables and their complex interactions being, thus, a valuable tool for feature engineering and the development of a data-driven approach to risk management.

--