Evaluating the Generalisability of Deep Learning Models for Global Fertility Rate Forecasting: A Region-Aware Comparative Analysis

Rodrigues Da Silva Junior, Jair (2025) Evaluating the Generalisability of Deep Learning Models for Global Fertility Rate Forecasting: A Region-Aware Comparative Analysis. Masters thesis, Dublin, National College of Ireland.

Preview	PDF (Master of Science) Download (726kB) | Preview
Preview	PDF (Configuration Manual) Download (406kB) | Preview

Abstract

Global fertility rates have declined considerably in recent decades, leading to ageing populations and shrinking workforces that challenge economic growth and strain healthcare and pension systems. Accurate forecasting of fertility rates is therefore essential for demographic planning and social policy design. This study evaluates the generalisability of seven deep learning architectures (Dense, LSTM, Auto-LSTM, Attention-based LSTM, GRU, BiLSTM, and CNN) using harmonised socioeconomic indicators from 2004 to 2023 across multiple world regions. Spain was used as a reference country for model evaluation and hyperparameter selection. In the Spain pilot, the non-temporal Dense network achieved the lowest prediction error, although several recurrent architectures performed competitively and offered temporal modelling capabilities. A recurrent architecture was therefore selected for cross-regional testing to assess the transferability of learned temporal fertility patterns. When applied to other countries, performance improved substantially under per-country feature rescaling, with weighted MAE reductions ranging from 53% to 97% across regions. The findings show that deep learning models can capture generalisable temporal fertility structures when supported by region-aware preprocessing, and illustrate the policy relevance of improved predictive accuracy: even a small percentage-point reduction in error can translate into thousands of correctly anticipated births in large populations, aiding planning in healthcare, education, and social protection systems.

Item Type:	Thesis (Masters)
Supervisors:	Name Email Hasanuzzaman, Mohammed UNSPECIFIED
Subjects:	H Social Sciences > HQ The family. Marriage. Woman R Medicine > Healthcare Industry Q Science > Q Science (General) > Self-organizing systems. Conscious automata > Machine learning
Divisions:	School of Computing > Master of Science in Data Analytics
Depositing User:	Ciara O'Brien
Date Deposited:	09 Sep 2026 08:20
Last Modified:	09 Sep 2026 08:20
URI:	https://norma.ncirl.ie/id/eprint/9902

Actions (login required)

View Item