Prezzaggio di polizze non-vita utilizzando il Data-Science: il caso francese

Starting from the data set of the Third Actuarial Pricing Game, containing the characteristics of 100,000 French policyholders, we have estimated the actuarial pure premium, conditioned on the characteristics of drivers, cars, and relevant policy in contracts using first the Generalized Linear Model (GLM) and the Generalized Additive Model (GAM). Then, we have extended these classical actuarial algorithms using some supervised learning methods to achieve optimal predictive power. In addition to that, we have represented the geographical distribution of the variables under study by exploiting the combination of longitude and latitude indicating the residence of the policyholders themselves.In the first chapter, we discuss the issues of non-life insurance focusing on the French Car Liability insurance regulation. In the second chapter, we discuss the theoretical properties of the models under study. In the third chapter, we apply the methods to the problem of predicting the claims frequency and severity.We conclude that the machine learning techniques lead to a better performance on the test set with respect to the classical models