Improved mortality rate forecasting using machine learning and open data
A multi-population approach to mortality rate forecasting using open data and interpretable neural networks.
Compared with traditional mortality models, machine learning algorithms can significantly improve the forecasts of future mortality rates.
Longevity is a major factor in the profitability of life insurers throughout the world. Mortality forecasts could therefore have a substantial impact on their financial results. In this paper, we investigate using open data and advanced modelling approaches to improve mortality forecasts. We illustrate the performance of the method on mortality for France and the Netherlands. We trained a Temporal Fusion Transformer (TFT) model on multi-population, age-specific, mortality data—enriched with socio-economic data collected by the World Bank. Our discussion includes:
- The TFT model
- Data used: The Human Mortality Database
- Our model: training, interpretation and evaluation