Uncertainty Quantification in Demographic and Health Forecasting Using Probabilistic Artificial Intelligence
Abstract: Reliable demographic and health forecasting requires accurate predictions and appropriate quantification of uncertainty. This study compares classical statistical models, supervised machine learning, deterministic deep learning, and probabilistic deep learning for forecasting life expectancy across 54 African countries from 2000 to 2023. The dataset contains 1,296 country-year observations and 25 candidate socioeconomic and public health predictors. Models are evaluated over one-, three-, and five-year forecasting horizons using point prediction accuracy, prediction interval coverage, and uncertainty measures. Results indicate that classical ARIMA achieves strong short-term accuracy, with a mean absolute error of 0.652 at the one-year horizon, while XGBoost provides the lowest five-year prediction error of 0.844. Probabilistic deep learning models using Monte Carlo Dropout demonstrate favorable uncertainty calibration, with the MC-Dropout multilayer perceptron achieving 96.3% prediction interval coverage at the five-year horizon. Its estimated epistemic variance share increases from 33.5% to 42.8% between the one- and five-year horizons. These findings highlight the importance of evaluating both predictive accuracy and uncertainty calibration when developing demographic forecasting systems. Probabilistic approaches can complement conventional forecasting models by providing additional information for risk-aware public health planning.
Keywords: Demographic Forecasting; Life Expectancy; Probabilistic Artificial Intelligence; Bayesian Deep Learning; Uncertainty Quantification; Monte Carlo Dropout
How to Cite?: Ridab Abdalla, Altaiyb Mohmmed, "Uncertainty Quantification in Demographic and Health Forecasting Using Probabilistic Artificial Intelligence", Volume 15 Issue 10, October 2026, International Journal of Science and Research (IJSR), Pages: 1424-1431, https://www.ijsr.net/getabstract.php?paperid=SR261018193709, DOI: https://dx.doi.org/10.21275/SR261018193709