Vyhodnocování pojistných událostí pomocí strojového učení

This bachelor thesis describes machine learning techiques, neural networks in particular and their application for insurance claims evaluation. One set of such techniques is then applied to a specific task of modeling the severity of insurance claims related to car accidents. The introductory part explains the motivation to model insurance claims and its importance in the creation of insurance reserves. The model example in this thesis uses data from the american insurance company AllState. This data was made available for an online predictive modelling competition held by the insurance company in 2016. In the applied section of this thesis, three different architectures of neural networks are presented and their performance is evaluated. The best neural network is then used for the final model ensemble. The ensemble achieves a result that is more accurate by 8% than the benchmark value submitted by the competition organizer. This thesis hence demonstrates the potential of machine learning applications in the insurance industry.