Improving geographically weighted Poisson regression model based on metaheuristic algorithms: Application to cancer rate data

Yousif, Hind Mohammed; Algamal, Zakaria

doi:10.21608/cjmss.2025.364215.1131

Improving geographically weighted Poisson regression model based on metaheuristic algorithms: Application to cancer rate data

Document Type : Original Article

Authors

Department of Statistics and Informatics, University of Mosul, 41002 Mosul, Iraq

10.21608/cjmss.2025.364215.1131

Abstract

Geographically weighted Poisson regression (GWPR) model is a further refinishing of Poisson regression for model the spatial count data and consider local association of variables. Nevertheless, the GWPR model faces several challenges that can impact its effectiveness and reliability. One of these challenges is the bandwidth selection. An improper bandwidth value results to either fitting the GWPR model to the noise or output values that are unexpectedly low. A small bandwidth may well include too much local variability, while large bandwidth may average out important local changes. Meta-heuristic algorithms can be defined as optimization methods that designs up approximate solutions to problems, which involves searching through the solution space in the best way possible. Employment of meta-heuristic algorithms in determining the bandwidth value in GWPR model is entirely novel owing to the utilization of optimization methods in the selection of bandwidth value. In this paper, beluga whale optimization algorithm as meta-heuristic algorithms are employed to find the best value of the GWPR model bandwidth by considering the objective function for bandwidth selection as minimizing prediction errors. Based on cancer rate estimation as a real data application, the comparison studies and evaluations demonstrated that the proposed method outperformed other methods regarding pseudo-R2 and Deviance. According to the results, utilizing the meta-heuristic algorithms for estimating the bandwidth value in GWPR model presents a promising approach that combines advanced optimization techniques with spatial analysis.

Keywords