THE USE OF KERNEL FUNCTION AND BANDWIDTH IN GEOGRAPHICALLY WEIGHTED REGRESSION (GWR) MODELING (CASE STUDY: GOLD PROSPECT DATA IN ACEH)

• Main Author: Erienci Fitri, Lucci
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/39120
• Institution: Institut Teknologi Bandung

Minerals such as gold has an important role in increasing state revenues. In Indonesia, there are several regions that have gold prospects, one of which is Aceh. The gold prospect data in Aceh is a gold finding data followed by findings of 34 other metal and mineral elements that have spatial relationship with gold. From prior research on the same data, a strong correlation between gold and silver and bismuth was found. This study aims to conduct further searches on that relationship using Geographically Weighted Regression (GWR). Some GWR models are formed with different types of kernel functions (Gaussian and bisquare) and different bandwidths (fixed and adaptive) using cross validation (CV) as bandwidth selection criteria. Then also coefficient of determination is used to determine the most appropriate model to represent data. The results show that GWR model using Gaussian kernel function adaptive bandwidth is better at modeling the relationship between gold and silver and bismuth than other kernel functions. Gaussian kernel function adaptive bandwidth produces the optimum bandwidth of 23 nearest neighbors. This means that the findings of gold in a location are influenced by 23 gold findings around it.