Enhancing COVID-19 analysis using adaptive robust geographically weighted regression: a global perspective

Abstract

Background: Understanding the global spread and impact of COVID-19 requires analytical approaches that capture spatial heterogeneity and data irregularities. Traditional regression methods often fail to address issues such as outliers and heteroscedasticity, limiting their effectiveness in modeling pandemic data.

Methods: This was an ecological and spatial analytical study using secondary global COVID-19 data. This study employs an adaptive robust geographically weighted regression (AR-GWR) model integrated with the γ-divergence technique to enhance robustness against outliers and non-constant variance. Unlike classical regression and standard geographically weighted regression (GWR), the AR-GWR model incorporates adaptive bandwidth selection, enabling automatic optimization of spatial smoothness and robustness parameters. This improves localized parameter estimation and predictive performance.

Results: The findings reveal substantial spatial variation in COVID-19 outcomes across countries. Nations with advanced healthcare systems, such as South Korea and France, report higher case numbers despite high human development index (HDI) scores. In contrast, Sub-Saharan African countries exhibit relatively lower case and mortality rates, potentially due to demographic and geographic factors. The AR-GWR model identifies healthcare infrastructure and preexisting health conditions as significant determinants of COVID-19 mortality, while demographic factors primarily influence infection rates. Compared to traditional models, AR-GWR demonstrates superior predictive accuracy and better handling of spatial non-stationarity.

Conclusions: The study highlighted the effectiveness of advanced spatial regression techniques, particularly AR-GWR with adaptive bandwidth selection, in modelling complex pandemic data. By accounting for spatial heterogeneity and data irregularities, the model provides more reliable insights into global health patterns. These findings can support improved pandemic preparedness and informed public health decision-making.