Robust Geographically Weighted Regression Modeling using Least Absolute Deviation and M-Estimator

Abstract

Geographically weighted regression (GWR) is development of multiple regression that has spatial varying, so that the estimator of GWR is different for each location. Parameter estimation in GWR uses weighted least square method which is vulnerable to outlier and can cause biased parameter estimation. The robust GWR (RGWR) with LAD and M-estimator is resistance to outliers. This research estimated parameters on RGWR using LAD and M-estimator method and uses data of Java gross domestic product (GRDP) in 2015 containing several outliers. The result showed that RGWR model was better than GWR with M-estimator, and the predictions were closer to the actual values.