Mortality Forecasting Using LSTM-CNN Model

This study proposed a deep neural network (DNN) model for the mortality rates of multiple populations, comprising the long short-term memory (LSTM) and convolutional neural network (CNN) models. As mortality trends are reflected for a long period and summarizing the complicated dependency of mortality rates across countries or regions using a linear model is challenging, LSTM and CNN were applied to mortality modeling. The former can automatically learn long-term dependencies of sequential data, whereas the latter can extract local features of grid or sequential data. Thereafter, the DNN and baseline models were used to fit mortality data of 21 countries from Human Mortality Database (HMD), which was divided into training and test sets considering the year 2000. Extensive numerical experiments from the perspective of accuracy stability and reliability of long-term forecasting revealed that DNN models yield an enhanced predictive performance, particularly the LSTM-CNN model. It combined LSTM, CNN, and FCN layers, and thus exploited each deep neural network to fit nonlinear age, period, and cohort effects as well as their interactive terms to provide the strongest predictive power. However, CNN was still predominant for certain groups. Furthermore, the conclusions also hold for remaining life expectancy