Construction of Mortality Tables using LSTM Neural Networks

Abstract

Mortality Table are tables structured with mortality database, especially mortality rates observed at all ages, used in pension funds and life insurance. This article based about application of neural network model to construction of forecast life tables comparing the performance of future model to Lee-Carter Model. Lee-Carter has been used in literature for fitting and forecasting the human mortality rates in mortality table. Architecture proposed was LSTM (Long-Short Term Memory) Network Neural model. LSTM Neural Networks are ideal for prediction of temporal sequences. Dates were collected through historical mortality tables information of IBGE (Instituto Brasileiro de Geografia e Estatistica). Results providing plausible utilily of model LSTM Neural Network to approach to forecasting mortality rates.