Kajian Metode Least Absolute Selection and Shrinkage Operator (LASSO) pada Data yang Mengandung Heteroskedastisitas

Abstract

Least Absolute Selection and Shrinkage Operator (LASSO) has been acknowledged to analyse high dimention data to select variables and to estimate parameters. LASSO estimators obtained by minimizing the residual sum of squares subject to the sum of the absolute value of the coefficients being less than a constant. Jia et al. (2010), in his research, conducted an analysis on a medical imaging application data using LASSO when error variance of the data suffered heteroscedasticity problem, which is Poisson-like distributed. This research aimed to study the similar problem. LASSO is evaluated by using heteroscedastic regression data. By conducting simulation approach, the result showed that LASSO encountered difficulties. In regression data that has too many zerocoefficients estimator, LASSO is not selective. Compared to OLS (Ordinary Least Square) and Best Subset, LASSO doesn’t offer better solution