Classical regression analysis with the OLS (ordinary least square) has several assumptions. One of the assumptions is that there is no multicollinearity in the predictor variables. If multicollinearity occurs in the data, there are several other methods that can be used, including lasso regression and ridge regression. These two regression models are shrinkage methods that can shrink the regression coefficient so that the variance decreases. In this study, the performance of ridge regression and lasso regression was compared for data with multicollinearity. The result of the mean of squared errors (MSE) shows that the performance of the ridge regression is better than the lasso regression. In terms of model interpretation, lasso regression is considered superior. This is because lasso regression can shrink some coefficients to zero so that only 4 of the 9 variables used in the final model.
Copyrights © 2022