NEIGHBOURING PREDICTION FOR MORTALITY

Abstract We propose a new neighbouring prediction model for mortality forecasting. For each rate at age x in year t , m x,t we construct an image of neighbourhood data around that is, Ꜫ ( 1 2 s ), which includes information ages [ - + ], lagging k years (1 ≤ ). Combined with the deep learning – convolutional neural network, this framework is able to capture intricate nonlinear structure data: effect, can go beyond directions period, age, and cohort as classic models. By performing extensive empirical analysis on all 41 countries regions Human Mortality Database, find proposed models achieve superior forecasting performance. This be further enhanced patterns interactions between multiple populations.