Using machine learning to predict loss events

V magistrskem delu je predstavljena uporaba posplošenega linearnega modela in različnih metod strojnega učenja v zavarovalništvu. Delo je razdeljeno na teoretični in praktični del. Na začetku teoretičnega dela so opisani osnovni pojmi iz verjetnosti in zavarovalništva. Predstavljeno je tudi, kako zavarovalnice določijo višino premije. Nato sta predstavljena teoretično ozadje posplošenega linearnega modela in uporaba tega modela za napovedovanje višine škode. Na koncu teoretičnega dela pa je opisano strojno učenje in bolj podrobno so predstavljena odločitvena drevesa, naključni gozdovi ter nevronske mreže. V praktičnem delu magistrskega dela pa so posplošeni linearni model, naključni gozd in nevronska mreža uporabljeni za napovedovanje višine škode pri avtomobilskem zavarovanju. Najprej so podatki predstavljeni ter ustrezno obdelani. Nato so določeni parametri posameznih modelov. Na koncu pa so modeli med seboj primerjani in izbran je najboljši model.The master thesis presents the use of generalized linear model and different machine learning methods in insurance. It is divided into theoretical and practical part. At the beginning of the theoretical part, basic notions of probability and insurance are described. It is also presented, how insurance companies determine the insurance premium. Then the theoretical background of generalized linear model and the use of it in predicting claim amount are presented. At the end of the theoretical part, machine learning is described and also decision trees, random forests and neural networks are presented in detail. The practical part of master thesis is focused on how generalized linear model, random forest and neural network are used for predicting car insurance claims. First, the data is presented and processed. Then the parameters of each model are determined. In the end, the models are compared and the best one is chosen