Prediction of the number of assistance cases based on the development of the insurance portfolio

The thesis deals with the prediction of the number of assistance cases in travel insurance using several machine learning models and their comparison. Furthermore, the work focuses on the estimation of external costs related to assistance cases. Several models for predicting the number of assistance cases are estimated in the thesis, namely a basic model using the average number of cases in the identified segments. Furthermore, a Generalized Linear Model was estimated, which is a common methodology used in the insurance industry to predict number of claims. Last but not least XGBOOST machine learning model is estimated. Average costs in individual segments of assistance cases were used to determine the amount of external costs or the severity of damages. By comparing the evaluation metrics RMSE, MSE and MAE, the model using the GLM methodology was identified as the most appropriate model providing the best predictions of the number of assistance cases. The best results were achieved by the model in which the distribution channel, the age group of the insured, the length of the insurance and the season enter as explanatory variables. This model is further used to estimate claims severity and total external costs.