Forecasting HIV Case Dynamics in Enugu State, Nigeria: An Empirical Study of Growth Models and Ensemble Learning Algorithms

International audience The accurate prediction of HIV cases in Enugu State is critical for effective public health planning. Early intervention remains very limited due to inconsistent fluctuations in daily reported cases and the inability of a single model to capture the structural changes in the epidemic. The study aims to develop a framework for forecasting HIV case dynamics in Enugu State, Nigeria, integrating growth curve models and Ensemble algorithms. Using 17 years of cumulative HIV dataset from 2007–2023, smoothed using 7-day rolling mean, four nonlinear growth models: Exponential, Gompertz, Logistic, and Richards were fitted. Structural breaks were detected using the PELT algorithm and validated using the Chow test. The models were all fitted within each segment of the data, and the best performing model was used for that segment. The predictions from each segment were combined using three ensemble techniques: weighted average, Random Forest, and Gradient Boosting. The weighted ensemble showed the highest accuracy with R2 of 0.9996 and RMSE of 104.62 as well as strong uncertainty performance (IFMS = 662.56; ICP = 0.92). The study concludes that segmented growth modeling with ensemble learning significantly enhances the accuracy, stability, and interpretability of HIV forecasts, informing public health policy with reliable data in Enugu State. The use of routine surveillance data that sometimes contain reporting inconsistencies and omission of external determinants of HIV transmission such as as socioeconomic conditions, behavioural factors, and healthcare policy changes are the limitations of the study. The integration of these covariates and evaluation of the framework across multiple regions to enhance generalizability can be pursued in the future studies.