Regularisasi Regresi Linier Berganda pada Data Berdimensi Tinggi Untuk Mengatasi Efek Multikolinearitas

Abstract

Regularization on high-dimensional data is performed to determine whether regularization can address the effects of multicollinearity that arise when conducting multiple linear regression on high-dimensional data. High-dimensional data is known for its susceptibility to multicollinearity due to the characteristic where the number of observed variables exceeds the number of observations (p ≫ n). Ridge regularization and the Least Absolute Shrinkage and Selection Operator (LASSO) are used with the evaluation metric Mean Squared Error (MSE). The analysis of the designed synthetic data revealed that LASSO regularized regression tends to provide better performance compared to conventional linear regression and Ridge regularized regression, based on the minimal MSE value. This MSE value indicates that LASSO regularized regression offers the best performance in mitigating the effects of multicollinearity in high-dimensional data.