Multiscale geographically weighted regression with LASSO and Group LASSO: Review and application to micro and small enterprises revenue

Micro and Small Enterprises (MSE) holds a crucial role in the economy because it contributes 55% of the state’s income, but MSE still has a lot of deficiencies, so immediate optimization is vital. The purpose of this study is to model and map the MSE income at the regency level in West Java using Multiscale Geographically Weighted Regression (MGWR) with a selection variable process. MGWR is a method that is used to capture a spatial heterogeneity process by allowing effects to vary over space using “borrowed” nearby data that is controlled by various bandwidths for each variable. This research also adds variable selection processes such as LASSO and Group LASSO as an improvement of MGWR to model group-structured data. The response of this study is MSE income in 27 regencies/cities in West Java province, Indonesia, with 144 independent variables that will be selected using LASSO and Group LASSO to become predictor variables in MGWR model. The results of the spatial modelling show that the best model is MGWR with Group LASSO using bi-square kernel function. Based on this result, it can be seen that a group of important variables which significantly affect the MSE income are fertility, energy source, natural disaster, industry, and tourism. Fertility and energy source significantly affect the MSE income in all regencies, but fertility itself has no significant effect in big cities. Then, in the industry and tourism, the number of visiting foreign tourists has the most significant effect.