Dipartimento di Scienze Statistiche

This paper focuses on the potential of machine learning tools in micro-level reserving by using individual claim data, which is more and more available nowadays. This is especially relevant for non-life insurance, but it could also be useful for some specific life business branches. After a brief introduction to the problem of reserve estimation in non-life, we will describe the algorithms behind some of the fundamental machine learning tools such as regression methods, naive Bayes, k-nearest neighbors, CARTs, and neural networks. All of them will be used to estimate closing delay and payment amount for individual claims of a specific automobile bodily injury claim dataset. Theoretically, these estimations represent the foundation for a triangle-free, machine-learning-based approach to non-life reserving