Veen, Jasper, van der (2025) Predicting fertility across 11 European countries. Master thesis, Sociology.
|
Text
Predicting-fertility-across-11-European-countries.pdf Download (1MB) | Preview |
Abstract
This thesis investigates the predictability of short-term fertility outcomes across European countries using three machine learning models: penalized logistic regression, support vector machines, and extreme gradient boosting (XGBoost). Drawing on harmonized international survey data, the analysis explores both theoretical and technical factors that may influence model performance, including sample size, class balance, and model-specific characteristics. To ensure robustness, all models are evaluated through repeated cross-validation and assessed using multiple performance metrics, including accuracy, ROC-AUC, F1 score, and Brier score. Findings indicate that predictive performance varies substantially across countries, with class distribution and model architecture playing a central role. XGBoost consistently outperforms the other models, particularly in countries with balanced class distributions. Variable importance analyses reveal that while fertility intentions are key predictors in linear models, age-related variables dominate in tree-based approaches. The study highlights the need to complement explanatory approaches with predictive frameworks and offers methodological insights for future research on demographic behaviour.
| Item Type: | Thesis (Master) |
|---|---|
| Supervisor name: | Stulp, G. and Huisman, J.M.E. |
| Degree programme: | Sociology |
| Differentiation route: | Politiek, Maatschappij & Beleid [Master Sociology] |
| Date Deposited: | 01 Oct 2025 11:49 |
| Last Modified: | 01 Oct 2025 11:49 |
| URI: | http://gmwpublic.studenttheses.ub.rug.nl/id/eprint/5955 |
Actions (login required)
 |
View Item |