@article{Zuhdi_Fatatik_Prihasno_Rozaq_2025, place={Purwokerto}, title={Geographically Weighted Random Forests for Human Development Index of Central Java Prediction}, volume={6}, url={https://jutif.if.unsoed.ac.id/index.php/jurnal/article/view/5204}, DOI={10.52436/1.jutif.2025.6.4.5204}, abstractNote={<p>The geographically weighted regression (GWR) model has been widely used in various types of predictions, including human development index predictions. Similarly, the random forests (RF) model has also been widely used in various value predictions. The GWR model always assumes a local linear relationship between dependent and independent variables. The RF model only produces one global model that cannot represent conditions at each location. The GWR model is susceptible to multicollinearity in each independent variable, which can lead to overfitting if multicollinearity in the model is high. To address the vulnerability of the GWR model to multicollinearity, the RF model and the GWR model can be combined. Since the RF model is not vulnerable to multicollinearity in the independent variables, the modification becomes the geographically weighted random forests (GWRF) model to improve the shortcomings of the GWR and RF models. The GWR and GWRF models were constructed using data from districts and cities in Central Java Province, which was selected as the study area due to evident disparities in human development index achievements. These disparities highlight the presence of spatial heterogeneity that conventional models fail to adequately capture. To rigorously evaluate model performance, data from 2023 were employed as training data, while data from 2024 served as testing data. This research introduces a novel integration of spatial econometric and machine learning approaches, providing a more robust framework for addressing complex spatial variations in human development outcomes. The GWRF model is capable of producing a model that does not overfit when there is multicollinearity among independent variables. The GWRF model offers a novel integration of machine learning and spatial modelling, outperforming both GWR and RF by not only delivering high predictive accuracy under complex variable relationships but also capturing nuanced local spatial heterogeneity that conventional approaches fail to address.</p>}, number={4}, journal={Jurnal Teknik Informatika (Jutif)}, author={Zuhdi, Shaifudin and Fatatik, Isna Nurul and Prihasno, Izlah Nur Fadlila Herawati and Rozaq, Hasri Akbar Awal}, year={2025}, month={Sep.}, pages={2756–2768} }