Perbandingan Regresi Lasso dan Principal Component Regression (PCR) dalam Mengatasi Masalah Multikolinearitas

Abstract

Multicollinearity is a problem that often arises in regression analysis. This condition occurs when the independent variables have a high correlation between each other. If the assumption of the absence of multicollinearity is not met, researchers will have difficulty in identifying independent variables that have a significant effect in the regression model. The presence of multicollinearity can cause the estimation of regression parameters in the Ordinary Least Square (OLS) method to be inefficient. To overcome this, the LASSO Regression method and the Principal Component Regression (PCR) method are used. The data used in this study are generation data derived from low (0,1-0,3), medium (0,4-0,6), and high (0,7-0,9) correlation levels with different sample sizes (n=20,40,120,200) from normal distribution with 30 and 60 independent variables. The performance of LASSO Regression method and Principal Component Regression (PCR) method is evaluated using Mean Square Error (MSE) value and coefficient of determination (R2)) value. Based on this research, the LASSO Regression method has better efficiency than the Principal Component Regression (PCR) method because the LASSO Regression method obtains a smaller Mean Square Error (MSE) value and a higher coefficient of determination (R2) value than the Principal Component Regression (PCR) method.