Understanding the spatial non-stationarity in the relationships between malaria incidence and environmental risk factors using Geographically Weighted Random Forest: a case study in Rwanda

As found in the literature on health studies, the levels of ecological association between epidemiological diseases have been found to vary across regions. Due to limited research, little is known about how spatial environmental factors influence the variability of malaria incidence at smaller scales. We implemented the geographically weighted random forest (GWRF) machine-learning algorithm to analyze ecological disease patterns caused by spatially non-stationary processes using a malaria incidence dataset as well as a suite of diverse resolution environmental covariates for Rwanda. We first compared the geographically weighted regression (GWR), the global random forest (GRF), and the geographically weighted random forest (GWRF) to examine the spatial non-stationarity in the non-linear relationships between malaria incidence and risk factors. We used the Gaussian areal Kriging model to disaggregate the malaria incidence at the local administrative cell level to understand the relationships at a fine scale since the model goodness-of-fit was not satisfactory to explain malaria incidence due to the limited number of sample values at the health centre catchment level. Our results show that in terms of the coefficients of determination and prediction accuracy, the GWRF model outperforms the GWR and GRF models. The coefficients of determination of the GWR (R-2), the GRF (R-2), and the GWRF (R-2) were 0.47, 0.76, and 0.79, respectively. The local R-2 showed that the GWRF algorithm had higher performance in explaining the spatial variations of the non-linear relationships between malaria and the underlying factors, which could have implications for supporting local initiatives formalaria elimination in Rwanda.