Geographically Weighted Regression Analysis of Spatial Heterogeneity for Suicide Mortality

Abstract

Suicide mortality has a great spatial variation affected by socioeconomic status such as urbanization, divorce, unemployment, domestic violence, mental illness, and dissatisfaction and should be modeled locally instead of globally. The spatial data are supplied by the application of geographically weighted regression (GWR) model which takes into account local association of variables. However, GWR model has a number of challenges that may affect its effectiveness and reliability. The bandwidth choice is one of these difficulties. When there is an inappropriate bandwidth value, it leads to either fitting the GWR model to values of noise or unexpectedly low values of output. Small bandwidth can perhaps contain excessive local variability and large bandwidth can average away significant local variations. The meta-heuristic algorithms may be explained as the optimization algorithms, which is the solution aimed at designing up the approximate solutions to problems which implies the search through the solution space in the most appropriate manner. The use of meta-heuristic algorithms in the computation of the bandwidth value in GWR model is purely new due to the application of optimization technique in computing the bandwidth value. This paper has used pelican optimization algorithm (POA) as a meta-heuristic algorithm to derive an optimal value of GWR model bandwidth based on the objective function to select the bandwidth (bandwidth minimization). According to the estimation of suicide mortality rate as a real data application, the comparison studies and assessments showed that the proposed method performed better compared to the other methods in terms of R2 and Deviance. Based on the findings, the meta-heuristic algorithms of estimating the bandwidth value of GWR model is a prospective strategy which embraces a combination of sophisticated optimization strategies alongside with analysis of space.