A. Debón, S. Haberman, G. Piscopo
Studies across various countries reveal gender differences in mortality rates and life expectancy by nation. The multi-population Lee-Carter models decompose mortality rates into age, time, and country components, providing valuable insights into mortality trends. This study explores the model's interpretability, highlighting its capacity to uncover underlying mortality patterns and forecast future trends. Additionally, we enhance the model by integrating machine learning techniques to account for residual patterns not captured by the traditional framework.
Using up-to-date statistical techniques and data from the Human Mortality Database, we apply advanced computational algorithms to improve the accuracy of mortality rate predictions. Through empirical validation and comparative analysis, we demonstrate that combining machine learning with the Lee-Carter approach improves the accuracy of mortality predictions, advancing the methodology for more robust and accurate modelling.
Keywords: Multi-population Lee-Carter models, interpretability, forecasting, machine learning
Scheduled
Biostatistics II
June 13, 2025 9:00 AM
MR 3