A Cohort Based Mortality Forecast in England and Wales with Application of K-Nearest Neighbor Classification for Causes of Death

Abstract

Continuous increase in overall life expectancy has brought forward the importance of mortality forecasting. Estimates in mortality trends have become extremely crucial and informative in various fields, including the planning and funding of health care systems, pension schemes as well as pricing annuity portfolios and reserves. Furthermore, while it is universally agreed that while age and period effects have significant effects onto mortality modelling, cohort effects have presented mixed results, mainly favorable towards the population of England and Wales. Underlying cause-of-death has also become a topic of interest in terms of mortality modelling, as it brings significance towards the forecasted mortality rates, but have yet to be proven or thoroughly understood. Hence, the aim of this study is to compare several popular mortality models using the population of England and Wales, as well as test the significance of cohort effects within the population. This study also intends to classify deaths according to their respective causes using the k-nearest neighbor algorithm, allowing possible assumptions to mortality data without cause-of-death in future studies.

 Keywords: Mortality Forecasting, Lee-Carter, Cohort Effects, K-Nearest Neighbor (KNN) Classification