How do you detect and fix multicollinearity?

Suppose you are working with a regression-style model—such as linear or logistic regression—and you suspect that the features are affected by multicollinearity.

1. Define multicollinearity and explain why it causes issues.
2. What methods would you use to detect it, including diagnostics, plots, or statistical measures?
3. In practice, how would you address it while still maintaining predictive accuracy and/or interpretability?
4. How does your approach change depending on whether the goal is inference (interpreting the coefficients) or purely prediction? Overview: This question tests knowledge of multicollinearity along with related skills in regression diagnostics, feature engineering, model interpretability, and the trade-offs between interpreting coefficients and maximizing predictive performance.