Institutional Repository of Vilnius University
Mašininio mokymosi modelių palyginimas išgyvenamumo prognazavime /
Abstract
dc:descriptionThis study investigated and compared machine learning algorithms of different architectures a decision tree ensemble (XGBoost) and a localized neural network (LocalGLMnet) to forecast the mortality probabilities of the Lithuanian population and evaluate the impact of macroeconomic factors. The forecasting quality and accuracy of the algorithms were measured using RMSE, MAE, RMSLE, and the coefficient of determination metrics, while Explainable AI instruments SHAP values and internal attention weights—were employed to deconstruct their decisionmaking logic. The results showed that the optimized baseline models achieve exceptionally high predictive accuracy (R2 > 0.99), with LocalGLMnet demonstrating the best global performance. However, the inclusion of additional macroeconomic variables highlighted the vulnerability of flexible neural networks under smallsample conditions: the LocalGLMnet model became destabilized and overfitted, whereas the structurally more rigid XGBoost maintained stability and slightly improved its accuracy. Analysis confirmed that mortality risk is overwhelmingly determined by demographic inertia (age and gender), while the added value of shortterm macroeconomic cycles to the forecast is minimal. Finally, it was established that the remaining algorithmic deviations in extreme old age stem from stochastic noise and natural data sparsity within the small Lithuanian population.
Degree
thesis:*- Grantor dc:publisher
- Institutional Repository of Vilnius University
- Year dc:date
- 2026
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Selivonec, Andrej,
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:ELABAETD308327718&prefLang=en_US