Elastic-net Regression

Creator
Creator
Seonglae Cho
Created
Created
2025 Jan 28 12:44
Editor
Edited
Edited
2025 May 19 15:49
Refs
Refs

A combination of L1-norm and L2-norm regularization

w=argminw1Ni=1N(w,xiyi)2+P(w)w^* = \arg\min_{w} \frac{1}{N} \sum_{i=1}^{N} \left( \langle w, x_i \rangle - y_i \right)^2 + P(w)P(w)=λ(αw1+(1α)w22)P(w)=\lambda\bigl(\alpha\|w\|_1 + (1-\alpha)\|w\|_2^2\bigr)
where 0<α10 < \alpha \le 1 and λ0\lambda \ge 0 , directly treats
Bias-Variance Trade-off
. It does not consider
Ridge regression
case (α=0\alpha = 0).
 
 
 
 
 
 
 
 

Recommendations