Kernel Ridge Regression (KRR)

Abstract

Kernel regression is more sensitive than traditional ordinary least squares regression, but is a discretization model. By the add-up sum of Gaussians, continuous variables are converted into discrete ones, otherwise discretized ones.

Another problem is that of increasing mathematical complexity with multidimensional data. However, the kernel trick is an efficient and less computationally-intensive way to transform data into high dimensions. A third problem, is that of data overfitting. It can, however, be corrected by regularization, where regression coefficients (b-values) are penalized to a lower level according to b_ridge = b/(1 + λ) where λ = shrinking factor.