Hybrid Format

Deep Learning for Longevity Projections. State of Art and Future Directions

Andrea Nigri
Laboratory of Digital and Computational Demography, October 19, 2022

Andrea Nigri from the University of Foggia provided new insights into the analysis and forecasting of human longevity using recent innovations in deep learning.

Abstract

The last 30 years have been an exciting time in the history of mortality forecasting. During this period, innovative methods have been constantly proposed, contributing to the ability to anticipate population dynamics. This heightened research activity is the result of the intense adoption of statistical methods, which have built solid foundations for the evolution of mortality forecasting. Nevertheless, forecasting future longevity is not straightforward, and accuracy depends on the particular situation or trends. Indeed, it may be years before new techniques can be fully evaluated; thus, the accuracy of population projections should be regularly tested to ensure evidence of improvement. Researchers appear to be more focused on making technical progress in methods than on minimizing prediction errors.

This presentation aims to provide new insights into the analysis and forecasting of human longevity using recent innovations in deep learning. We will trace the short history of deep learning in demographic research up to the most recent contributions and attempt to disentangle how deep learning may help demographers. Therefore, we address whether there is still room for collaboration between deep learning and demography with a specific discussion about longevity forecasting.

About

Andrea is a Tenure-track Assistant Professor at the Department of Economics, Management and Territory, University of Foggia. Prior to joining the University of Foggia, he was a Research Fellow at the Department of Social and Political Sciences, Bocconi University. He works on developing and improving the statistical methodology in the field of Statistical Demography. He has an interdisciplinary background with training in Statistics and its application to demographic research, obtained through a Ph.D. at Sapienza University of Rome, Department of Statistics, and through the European Doctoral School of Demography.

His research has served as a bridge between Longevity analysis and Statistical Learning, introducing Regression trees and Neural Networks as integration or full replacement of traditional mortality models. His current research agenda contains two substantial main strands concerning the indirect estimation of vital rates from summary demographic measures, and causes-of-death modeling.