Abstract

Third-party liability is the most important sub-category in automobile insurance, that is why actuaries seek always to design the ideal price list by classifying the insureds into homogeneous classes. However, the heterogeneity persists in the priori tariffication. For that, actuaries use the Bonus Malus system to redistribute the cost of claims more equitably between insureds by rewarding good insureds with a bonus and penalizing bad insureds with a Malus. Nevertheless, the classical approach used in the conception of Bonus Malus systems is limited to the parametric methods that need to make the hypothesis of the number of claims distribution and don’t consider the cost of claims. In this direction, this paper seeks to avoid this issue by using machine learning algorithms, in response to offering a fair Bonus Malus System. Two models of posteriori tariffication will be built. In addition, three algorithms will be used, in occurrence, the CART Classification And Regression Tree method, SVM the Support Vector Machine for regression, and KNN the K-Nearest Neighbor. The suggested models take into account, not only the number of claims but also the importance of the cost of claims. A numerical illustration shows the flexibility of posteriori premiums calculated by our models in relation to the risk levels. This work is a start for new actuarial research which seeks to use artificial intelligence in the design of bonus malus systems.