Explain why LASSO selects features

Explain what makes LASSO perform feature selection. Include: 1) a high-level intuition contrasting L1 and L2 penalties; 2) a geometric reading of the constraint region and why coefficients become exactly zero; 3) the KKT/subgradient condition that determines when a coefficient is zero; 4) how correlated predictors affect selection stability; 5) why standardization is important and the consequence of skipping it; 6) how lambda is selected and how it moves the bias–variance trade-off; 7) when Elastic Net is preferred and the reason.

The question tests grasp of regularization and feature selection in linear models, spanning LASSO's L1 penalty compared with L2, geometric intuition for the constraint region, optimality/KKT conditions, correlated-predictor effects, the role of standardization, hyperparameter choice, and when Elastic Net is suitable, in the Machine Learning area for Data Scientist positions. It appears often because it checks both conceptual knowledge and practical use of model sparsity, interpretability, preprocessing, and bias–variance trade-offs, assessing statistical optimization and model selection rather than implementation specifics.