Hybrid Probabilistic Forecasting of Under-Five Malaria Admissions in Ghana: A Gaussian Process Regression with Holt-Winters Smoothing
Title: Combining Gaussian Process Regression with Holt-Winters Smoothing for Probabilistic Prediction of Under-Five Malaria Hospitalizations in Ghana
Abstract:
In sub-Saharan Africa, precise malaria forecasting is hindered by significant obstacles, including pronounced seasonal variations, inconsistencies in data reporting, and non-stationary transmission patterns, all of which compromise the effectiveness of standard modeling techniques. Within Ghana, the necessity for district-level malaria surveillance demands forecasting systems that are statistically rigorous and resilient, even when working with constrained datasets. To address this, the present research introduces a novel hybrid methodology that merges Gaussian Process Regression (GPR) with Holt-Winters exponential smoothing to model monthly admissions for malaria among children under five. This approach leverages GPR to account for non-linear trends and quantify predictive uncertainty, while simultaneously employing Holt-Winters smoothing to stabilize long-term projections and maintain the integrity of seasonal cycles.
The study utilized a decade of district-specific data spanning from 2014 to 2023. Model performance was assessed using a rolling-origin expanding-window validation technique. Results demonstrated that the hybrid model significantly outperformed the standalone Holt-Winters approach, achieving an $R^2$ of 0.9906 compared to 0.8213. Furthermore, 94.2% of the model's residuals fell within the $\pm 2\sigma$ confidence intervals. Looking ahead, projections for the period 2024–2028 estimate that average monthly admissions will range between approximately 8,000 and 12,200 cases.
A spatio-temporal examination uncovered distinct ecological disparities: while northern districts with high disease burdens experienced substantial absolute variations in case numbers, their relative trends remained consistent. Ultimately, this framework offers a scalable, probabilistic tool for early warning systems and operational planning in endemic regions, thereby bolstering Ghana’s national efforts to control malaria.
Source: arXiv Generated at: 2026-06-02 00:00:00 UTC




