Determination of the best multivariate adaptive geographically weighted generalized Poisson regression splines model employing generalized cross-validation in dengue fever cases

• Published in: MethodsX 2023-01, Vol.10, p.102174-102174, Article 102174
• Main Authors: Sriningsih, Riry, Otok, Bambang Widjanarko, Sutikno
• Format: Article
• Language: English
• Subjects: DHF; GCV; Generalized Poisson; GWGPR; MAGPRS; MAGWGPRS; MARS; MSE; Multivariate Adaptive Geographically Weighted Generalized Poisson Regression Splines (MAGWGPRS); Review; Weighted-MLE
• ISSN: 2215-0161; 2215-0161
• DOI: 10.1016/j.mex.2023.102174

•We present MAGWGPRS, a new model that is a combination and development of count response MARS and spatial regression.•Selecting the optimal bandwidth using an adaptive Gaussian kernel function based on the CV method.•Grouping districts/cities in Java Indonesia based on inter-regional disaggregating variables in modeling DHF cases. This article constructs a new model based on multivariate adaptive generalized Poisson regression splines (MAGPRS) and geographically weighted generalized Poisson regression (GWGPR), which is known as multivariate adaptive geographically weighted generalized Poisson regression splines (MAGWGPRS). The article elaborates the steps of weighted maximum likelihood estimation (weighted-MLE) to obtain the estimated values of its parameters. MAGWGPRS and MAGPRS were applied to the number of dengue hemorrhagic fever (DHF) cases in 119 districts or cities in Java, Indonesia, in 2020, to compare their performance. The fitted value plot versus actual data and a comparison of the mean square error (MSE) value demonstrate the goodness of the two models. The best MAGWGPRS model for each location was obtained, and only one the best MAGPRS model for all locations was acquired. Based on the plot results of the fitted value with the actual data and MSE value, MAGWGPRS is determined to be superior to MAGPRS. [Display omitted]