Institutional Repository of Vilnius University
Sveikatos draudimo portfelio analizė ir rizikingiausių segmentų identifikavimas /
Abstract
dc:descriptionThis paper analyzes health insurance portfolio data to assess the applicability of different models for predicting pure insurance premiums. The study compares several forecasting methods and aims to identify links between risk factors and pure premiums in different portfolio segments. The variables selected for analysis are the insured age range, the share of insured employees in the company, activity category, location, and risk exposure. Three main models were created and compared in the study: GLM Tweedie, GAM Tweedie, and XGBoost Tweedie. Models based on the Gamma and inverse-Gaussian distributions were also evaluated, but due to poorer performance, were not included in the main analysis. The Tweedie distribution was chosen as the most suitable for insurance data. It allows modeling of asymmetric, highly dispersed data in which a large proportion of the values are equal to zero. The quality of the models was assessed using forecasting accuracy metrics, and information criteria were additionally analyzed for the GLM and GAM models. For model interpretation, coefficient estimates were analyzed for GLM and GAM, and variable importance was assessed for the XGBoost model. Results show that pure insurance premiums depend on several risk factors, whose impacts vary across portfolio segments. Comparing models revealed that in this case the XGBoost Tweedie model had the best forecasting accuracy. This model achieved the most optimal metric values and the highest determination coefficient. The study confirms that Tweedie-based models can be used for pure premium forecasting, and combining statistical and machine learning methods improves accuracy and assessment of risk factors.
Degree
thesis:*- Grantor dc:publisher
- Institutional Repository of Vilnius University
- Year dc:date
- 2026
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Proškinaitė, Vitalija,
Rights
dc:rights- Statement dc:rights
-
- info:eu-repo/semantics/openAccess
- Language dc:language
- lit
Identifiers
dc:identifier.*- Repository record dc:identifier
- https://repository.vu.lt/VU:ELABAETD308333275&prefLang=en_US