A multi-population approach to mortality rate forecasting using open data and interpretable neural networks.
Since the mid 20th century, human mortality has declined globally. In most industrialized countries, mortality among adults shows a decreasing annual probability of death. In the literature on mortality modelling, machine learning approaches have recently emerged. We propose in this article to predict future mortality from past data and to compare the output predictions of classic stochastic mortality models (baseline models) with those of machine learning model approaches (neural network and random forest). The article includes discussion of the following:
- The human mortality database
- Construction of the dataset
- Baseline models
- Machine learning approaches
- Comparison of model performances