Abstract

Reliable mortality statistics are essential for population monitoring, health policy, and progress toward the Sustainable Development Goals. Yet, many low and middle-income countries still face challenges in producing complete life tables due to limited coverage of civil registration and vital statistics (CRVS) systems. The reliance on abridged life tables (aggregated into five-year age groups) restricts the capacity of national statistical offices (NSOs) to provide accurate single-year estimates of mortality, particularly at early and older ages where risks are more volatile. This study introduces a hybrid methodology that integrates deterministic interpolation, deep learning models, and COM-Poisson smoothing to transform abridged life tables into complete single-age tables with greater accuracy and stability.
The approach is tested using Indonesian abridged life table data. First, interpolation methods (PCHIP, Spline, Akima, and Lagrange) are applied to disaggregate grouped death rates. Next, a deep learning model is trained to refine the age-specific mortality curve, capturing nonlinear patterns often missed by traditional graduation techniques. Finally, a COM-Poisson smoothing procedure is applied to stabilize the estimates, allowing transformation back into complete qx and lx functions. The results are benchmarked against official abridged tables and assessed using RMSE and MAE, with separate evaluation by sex.
Findings show that the hybrid model substantially improves accuracy, reducing error margins particularly in childhood and advanced ages. By combining machine learning with a flexible count distribution, the method outperforms conventional graduation models and provides smoother, more realistic mortality curves. Importantly, the framework is replicable in contexts where NSOs have only abridged or incomplete data. It offers a pathway for strengthening official mortality statistics, improving comparability across regions, and supporting CRVS-to-register transitions.
The contribution of this work lies not only in methodological innovation but also in policy relevance. By equipping NSOs with a practical tool to enhance mortality estimation under data limitations, the study responds directly to the demand for innovative, impact-driven statistical methods in the data revolution era. This aligns with the IAOS 2026 theme “Navigating the Data Revolution: Innovations and Impact in Modern Statistics”, and particularly with the strands on New sources and tools for effectiveness: AI and Machine Learning in Statistics and Data Interoperability.