With this thesis, we propose to evaluate the vehicle’s features in order to understand which ones are more related to risk from a motor insurance perspective. The variables from the entire portfolio are also studied in a similar way, in order to understand if there are other non-vehicle-related features that can help improve the pricing models. The used models for this were generalized linear models, having used two different approaches, namely a variable selection approach and a dimensionality reduction approach. Both these approaches are compared, including the advantages, disadvantages, as well as the results. Graphical models were also used in the portfolio risk assessment part, in order to explore the dependency structure of this dataset and its implications on the insurance side. In terms of structure, the thesis starts with an introduction, followed by a chapter regarding the data preparation and data description, and then by the mathematical background of the different methods used. Finally, we present the results obtained and their analysis, finishing with a conclusion.     «

With this thesis, we propose to evaluate the vehicle’s features in order to understand which ones are more related to risk from a motor insurance perspective. The variables from the entire portfolio are also studied in a similar way, in order to understand if there are other non-vehicle-related features that can help improve the pricing models. The used models for this were generalized linear models, having used two different approaches, namely a variable selection approach and a dimensionality...     »