Geographically Weighted Regression Model with Kernel Gaussian and Kernel Bi-square Weighting for Poverty Data (Case of 35 Villages in Jember Regency).

Abstract

Detection of the rural poverty, usually using the average expenditure per capita with a global analysis and the results are applied to all villages. But poverty is very likely influenced by region (space) and neighborhood, so the data between observation difficult to take independent. Geographically Weighted Regression (GWR) is an analysis to accommodate the spatial by involving geographical weighted, which is weighted geographically. Observations on the location of more weight with a smaller weight, according to Tobler's first law of geography which states that the location near the effect will be even greater. In this paper, the weighting used for the GWR model is Gaussian kernel function and bi-square kernel, with their each bandwidth values respectively. Optimal bandwidth can be obtained by minimizing the value of cross validation coefficient (CV). The results showed that the GWR model is more effective than the global regression. According to the mean square error (MSE) values, Gaussian kernel function is better than bi-square kernel as GWR weighting to analyze the data on average expenditure per capita of 35 villages in Jember Regency.