BERKALA SAINSTEK

Geographically Weighted Random Forest Modeling of Unemployment Rates in East Java Province: A Comparison of Fixed and Adaptive Kernels
(Pemodelan Geographically Weighted Random Forest pada Tingkat Pengangguran di Provinsi Jawa Timur: Perbandingan Fixed dan Adaptive Kernel)

Tiara Candra Mirella

Faculty of Mathematics and Natural Sciences, Universitas Jember, Jember, Indonesia

Yuliani Setia Dewi

Faculty of Mathematics and Natural Sciences, Universitas Jember, Jember, Indonesia

DOI: https://doi.org/10.19184/bst.v14i3.60068

ABSTRACT

Unemployment rates vary across regions due to differences in socioeconomic characteristics, indicating the presence of spatial heterogeneity. This study applies Geographically Weighted Random Forest (GWRF) to model unemployment rates across regencies and cities in East Java Province and to identify spatially varying variable importance. Two kernel approaches, namely fixed and adaptive kernels, are considered and compared based on their predictive performance. The model uses one dependent variable, unemployment rate, and six independent variables representing socioeconomic characteristics. Spatial heterogeneity is first examined using the Breusch–Pagan test. The GWRF model parameters are optimized through the specified tuning procedure, while bandwidths are selected separately for the fixed and adaptive kernels. Variable importance is evaluated using the percentage increase in mean squared error (%IncMSE). The results show that the adaptive kernel provides lower RMSE and MAE and higher R² than the fixed kernel. The adaptive kernel produces an RMSE of 0.1770565, MAE of 0.1363311, and R² of 0.9919788, whereas the fixed kernel produces an RMSE of 0.3487721, MAE of 0.249317, and R² of 0.9688756. The results indicate that the adaptive kernel provides better predictive performance under the evaluation procedure used in this study. The variable-importance analysis also indicates spatial variation in the relative importance of unemployment predictors across East Java Province.

Keywords: Unemployment rate; Geographically weighted random forest; Adaptive kernel; Fixed kernel.

PDF

Published

18-09-2026

Issue

Vol. 14 No. 3 2026: BERKALA SAINSTEK

Pages

161-170

License

Page updated

Report abuse