Probabilistic forecasting of under-five mortality in Uganda: implications for monitoring SDG 3.2

Tracking Sustainable Development Goal (SDG) 3.2 progress requires both explicit quantification of uncertainty and point estimates. Deterministic trend analysis, which has poor predictive power for reaching SDG targets, is the basis for most country-specific estimates of under-five mortality.

Probabilistic time series forecasting was used to analyse Uganda's annual under-five mortality data. Forecasts were created using Autoregressive Fractionally Integrated Moving Average (ARFIMA) and Autoregressive Integrated Moving Average (ARIMA) models through 2030. Rolling-origin root mean square error, Taylor diagram and residual diagnostics calibration were used to evaluate predictive accuracy. SDG 3.2-anchored metrics were computed, including the likelihood of achieving these goals by 2030, the absolute gaps that still exist, and the required annual rates of decline.

The ARIMA model predicts sustained declines in under-five mortality based on the available data but remains well above the target of 25 deaths per 1,000 live births set in SDG 3.2 by 2030. The model estimates a mortality gap of about 17.68 deaths per 1,000 live births, corresponding to a 1.98% probability of reaching the SDG target by 2030. Overall, these results suggest that the current rate of decline is insufficient to meet the target, and an average annual reduction of approximately 10.3% is necessary to achieve SDG 3.2. Even though the ARFIMA model showed moderate long-memory dependence, it produced substantially higher forecast errors and less stable long-term projections than the ARIMA model. Consequently, the ARIMA model was identified as the more accurate and reliable approach to predict under-five mortality in Uganda.

Despite realised values falling within anticipated uncertainty limits, Uganda is not on track to meet the SDG 3.2 under-five mortality target for 2030 based on current trends. A clear and practical method for monitoring the SDGs while accounting for uncertainty is probabilistic forecasting.