Perbandingan Pemodelan Geographically Weighted Lasso dan Geographically Weighted Regression Pada Kasus Kriminalitas di Jawa Timur

Abstract

Crime is a problem that is always faced and difficult to avoid in various countries, both developed and developing countries. Each district/city in East Java has differences in the distribution of crime cases. This is expected to have an impact on spatial dependence. Spatial data modeling takes into account spatial effects in the form of spatial heterogeneity. One of the spatial analysis methods is Geographically Weighted Regression (GWR) and there is multicollinearity in the data. Geographically Weighted Lasso (GWL) is used to overcome multicollinearity in crime cases in districts/cities in East Java. The aim of this research is to find out what factors influence the Geographically Weighted Lasso (GWL) and to carry out the best modeling of the GWR and GWL methods. GWL is a technique using the lasso approach in the GWR model to overcome multicolinearity. Comparasion between GWR and GWL uses RMSE and R-Square (R2) values. The RMSE value of the GWR model > GWL model ( 427,3693 > 194,4692 ). The R2 values of the GWR model is 0,67335, while the GWL is 0,71619 . Based on these values, it is conclude that the GWL model has better accuracy in explaining crime rate in East Java.