ISSN Online 3025-1605 Journal of Statistical Methods and Data Science Volume 01 Number 2 December 2023 https://jurnalmipa.unri.ac.id/jsmds Journal of Statistical Methods and Data Science 01(2) December 2023:44-51 44 IMPLEMENTATION OF GEOGRAPHICALLY WEIGHTED LASSO (GWL) IN ANALYZING RICE PRODUCTION FACTORS IN INDONESIA Reka Agustia Astari 1*, , Megawati 2, , Setyo Wahyudi 3. 1,2,3 Department of Statistics, IPB University, Jawa Barat, 16680, Indonesia *e-mail: rekaagustiaastari@apps.ipb.ac.id Article Info: Received: 25-03-2024 Accepted: 04-04-2024 Available Online: 05-04-2024 Keywords: global collinearity geographically weighted regression geographically weighted lasso spatial heterogeneity Abstract: Geographically Weighted Lasso (GWL) is a combination of two regression methods, namely Geographically Weighted Regression (GWR) and Least Absolute Shrinkage Selection Operator (LASSO). Both methods have their own uses. GWR is a regression that takes into account the geographical location aspect because the spatial heterogeneity test is not met. LASSO is a regression method to overcome multicollinearity in the data. The two problems are simultaneously contained in one regression model, namely the GWL method. This study will analyze the factors that affect rice production in 34 provinces in Indonesia by applying and interpreting the results of the Geographically Weighted Lasso method. The results of the analysis show that the coefficient of determination of the GWL model is 0.9703 so it can be concluded that the explanatory variables in this study can that the global level of rice production in each province in Indonesia is 97.03%. 1. INTRODUCTION Statistical methods are often used as a tool to determine the relationship between variables by forming a model that is appropriate in describing the characteristics of the data. As in linear regression models that are able to describe the relationship between explanatory variables and response variables. Looking at the relationship between variables in spatial data can be done with spatial statistics methods. Spatial data is geographically oriented data and has a certain coordinate system as its reference base, so that it can be presented in a map (Yulita, 2016). The problem that is often found in spatial data is the variety that is not always homogeneous at each observation location or called spatial heterogeneity. Spatial heterogeneity can be caused by several things such as differences in geographical conditions, socio-culture, and economic policies that vary in each location. This will be a problem if spatial data is still analyzed using the Least Squares Method (LSM) in estimating its parameters, because it can cause the variance of the estimates to be large. To overcome this problem, a method is needed that is able to overcome the heterogeneity of variance in spatial data to form a more efficient model (Yulita, 2016). Several previous studies on Geographically Weighted Regression (GWR) have been conducted including those conducted by Setiyorini et al. (2017) in their research on poverty in Java concluded that the Geographically Weighted Lasso (GWL) method is better than the GWR method on spatial data containing multicolinierity. Furthermore, the geographical layout of a region will produce different modeling, this is because differences in geographical location will affect the potential owned or used by a region (Pamungkas et al., 2016).