Yüksek boyutlu verilerde ridge, LASSO ve elastik net yöntemleri

Penalized linear regression methods that yield interpretable models with high prediction accuracy have gained great importance with the emergence of big and high-dimensional data sets recently. In this study, firstly, the ridge, LASSO, two-parameter estimation and elastic net methods which are popular methods in the literature are discussed in the context of penalized linear regression methods. In the next stage, the GO estimator is proposed as an alternative to these methods in case there is multicollinearity in the data set. After that, the shift-ridge and shift-LASSO estimators are proposed which can be used in modeling if the data set is high-dimensional in addition to having multicollinearity. Finally, the stochastic restricted LASSO estimator is proposed, which can be used as an alternative estimator if there are convenient stochastic restrictions for the data set. Real data set analysis and simulation studies are performed for each estimator. The proposed estimators are compared with other estimators in the literature by making use of C++ and R programming languages according to criteria such as test mean squared error and number of variables selected.