COMPARISON OF LASSO AND ADAPTIVE LASSO METHODS IN IDENTIFYING VARIABLES AFFECTING POPULATION EXPENDITURE

Abstract

Abstract: Since 2019, the difference in the increase in per capita expenditure of the population has continued to decline, and the most significant was only IDR 18,464 in 2021, indicating that the level of consumption of the population has not improved significantly, and the turnover of the community's economy is also not good. Multiple linear regression is more appropriate than other types of linear regression because it considers the influence of more than one independent variable on the dependent variable. However, problems may arise in the use of multiple linear regression, such as multicollinearity. To overcome this problem, other methods such as LASSO and adaptive LASSO should be used. Both methods have the ability to overcome multicollinearity between independent variables, thereby reducing the risk of misestimation. Nevertheless, the LASSO and Adaptive LASSO methods have differences in selecting important variables, so it is necessary to compare which method is better in terms of identifying influential variables. Based on the MSE and R-square comparison values, it is concluded that the Adaptive LASSO method model is the best model with a lower MSE value and a higher R-square value of 93%. The variable selection results of the Adaptive LASSO model are population size, number of households, average number of household members, constant price GDP, confirmed cases of COVID-19, human development index, percentage of the poor population, university student participation rate, and open unemployment rate.

GS Cited Analysis