Abstract
<p>Deep learning is transforming fields like chemistry, engineering, and medicine, but its application to social science fields like demography remains limited. This study explores deep learning as an emerging approach to mortality forecasting, a core area in demographic research. This paper addresses why and how deep learning performs better than three well-known baselines for mortality forecasting. First, it explores the relationship between model architecture and data pooling. Second, it provides a detailed country-level performance analysis, identifying contexts where deep learning underperforms and exploring possible remedies. Third, it discusses the broader applicability and potential of deep learning methods in demography. The findings underscore the potential of deep learning to enhance demographic forecasting and invite broader reflection on the role of computational and AI-driven methods in population studies.</p>