| Abstrakt: |
Spatial data is data that contains the influence of location with non-homogeneous variance at each location, or spatial heterogeneity. To address spatial heterogeneity, the Geographically Weighted Regression (GWR) model is used. However, in the GWR model, there is a phenomenon of multicollinearity, which is a strong relationship between independent variables that will reduce the accuracy of parameter estimation. To overcome multicollinearity in the GWR model, the Least Absolute Shrinkage and Selection Operator (LASSO) method is used. The LASSO method estimates the parameters of the GWR model by minimizing the sum of squared errors subject to a constraint function, which is solved using the Least Angle Regression (LARS) algorithm. This results in the Least Absolute Shrinkage and Selection Operator (LASSO) regression model to address the problem of multicollinearity in spatial data. Based on the research results, the LASSO method can overcome multicollinearity by shrinking the coefficients of parameters that contribute less and have a strong correlation with other independent variables in the GWR model, resulting in 33 final models. One of the models is for Nias Regency, where the factors influencing the poverty rate are the open unemployment rate, life expectancy, average length of schooling, gross participation rate, and per capita income. In Nias Regency, the value of s is 0.288 with an R-squared value of 0.9403. In Nias Regency, 94.03% of the variation in the poverty rate is explained by the independent variables in the model, while the remaining 5.97% is attributed to external factors not covered by the model. Coefficient of the Human Development Index variable shrinks to exactly zero, indicating that it has no effect on the poverty rate in Nias Regency. [ABSTRACT FROM AUTHOR] |
