ElasticNet regression is a combination of ridge and LASSO methods: add both penalty terms to the usual least squares loss function and you will get the ElasticNet regression. It also has two shrinkage parameters:
It is especially helpful when you have multiple correlated features. When two features are correlated, LASSO tends to choose one of them randomly, while ElasticNet keeps both. Similar to the ridge regression, ElasticNet is also more stable in many cases:
Figure 6.7: Penalty terms in the space of the model parameters. ...
