Comparison of GWANN and GWNBR in the Analysis of Positive Covid-19 Factors

Abstract

Covid-19 emerged with a rapid transmission rate, yet the precise factors contributing to its spread were initially unclear. To reduce the number of positive Covid-19 cases in East Java, spatial analysis is essential for examining these influencing factors. This study employs local research approaches, utilizing adaptive bisquare kernel weights. The local research, where the estimated value differs by region, is carried out using the Geographically Weighted Negative Binomial Regression (GWNBR) and the Geographically Weighted Artificial Neural Network (GWANN) method. This research aims to compare the GWANN, and GWNBR methods to determine the more effective approach for analyzing Covid-19 factors in East Java. The best model is selected based on the R² and RMSE values. The findings indicate that the Geographically Weighted Negative Binomial Regression (GWNBR) method is the more effective model, with an RMSE of 936.2055 and an R² of 0.902, and subsequent clustering based on significant local coefficients produced four regional groups; notably, comorbidity prevalence was significant in all areas, while other determinants differed by locality. These findings indicate that accounting jointly for overdispersion and spatially varying relationships improves inference on Covid-19 case patterns and highlights comorbidity burden as a province-wide priority for intervention.