A Gated Recurrent Unit Aided Lee-Carter Model for mortality projection

Abstract

Over the past decades, new models for mortality modelling have been developed. A
pioneering and one of the most influential model, among others, is the Lee-Carter
(LC) model, developed by Lee and Carter in 1992. The method traditionally used in
the LC model shows obvious drawback in describing the shape of future mortality.
This work used a more robust approach based on a deep learning technique that
differs from the literature. More precisely, a Gated recurrent unit (GRU) was
integrated with the Lee–Carter model to increase its capacity for prediction. The LC-
GRU model performs well in terms of predicted accuracy. The strength of the gated
recurrent unit network lies in its capability to handle the non-linearity inherent in
mortality data and long-term dependencies in time series data. The research results
shows that the LC-GRU model has minimum errors and thus performs better than
the LC model.