Compare How Models Respond to Multicollinearity

# Compare How Models Respond to Multicollinearity Compare multicollinearity in unregularized linear regression, L1-regularized linear regression, a decision tree, logistic regression, and gradient-boosted trees. A multiple-choice question asks which models are “not affected.” Explain which group is usually intended and why that wording is too absolute. ### What a Strong Answer Covers - Coefficient identifiability and variance problems in linear and logistic models. - What L1 regularization changes and why correlated predictors can still make selection unstable. - Why tree models avoid matrix-inversion identifiability problems while retaining other sensitivities. ```hint Separate prediction from interpretation Two interchangeable predictors can leave predictions similar while changing coefficients or feature importance substantially. ``` ### Follow-up Questions - Can Lasso select different predictors from the same correlated group across samples? - What can correlated features do to a tree model’s feature-importance scores?