Uncertainty‐Aware and Explainable Machine‐Learning Forecasts of Ghana’s Age‐Standardized HIV Incidence Toward 2030

Introduction Assessing progress toward ending the HIV epidemic by 2030 (SDG 3.3) requires forecasting frameworks that are accurate, externally valid, interpretable, and uncertainty aware. In Ghana, existing projections often rely on prespecified models with limited validation and incomplete treatment of epidemiological uncertainty. This study aimed to develop and apply a robust, explainable forecasting framework to evaluate historical dynamics and project Ghana’s age‐standardized HIV incidence rate toward 2030. Methods Annual age‐standardized HIV incidence rates for Ghana (1990–2023), with associated uncertainty bounds, were obtained from the Global Burden of Disease 2023 repository. A four‐stage framework was implemented: (i) comparative evaluation of eight statistical and machine‐learning forecasting models using expanding training windows and uncertainty‐aware metrics; (ii) optimization and locking of the best‐performing model via rolling‐origin cross‐validation; (iii) strict external validation on an untouched 8‐year holdout (2016–2023); and (iv) explainable machine‐learning (Shapley Additive exPlanations SHAP) analysis and recursive forecasting toward 2030 with Monte Carlo–derived prediction intervals. Results Extreme Gradient Boosting (XGBoost) consistently outperformed competing models and remained stable across training windows. On external validation, the tuned XGBoost model achieved a mean absolute error (MAE ≈ 896 incidence cases per year) and root mean squared error (RMSE ≈ 1067 incidence cases per year), and a mean absolute percentage error below 10%, with full empirical coverage of 80% and 95% prediction intervals. SHAP analysis identified short‐term incidence persistence and long‐term temporal trends as dominant drivers. Projections to 2030 indicated a largely stable trajectory, with a mean projected change of approximately +1.3% relative to 2023 and wide uncertainty spanning moderate decline and increase scenarios. Conclusion Current projections indicate that Ghana’s age‐standardized HIV incidence rate is unlikely to decline substantially by 2030, instead tending toward stabilization. This anticipated plateau underscores the necessity for intensified and targeted interventions to achieve meaningful reductions in new HIV infections within the Sustainable Development Goals timeframe.