Linear regression is not an outdated method. It is the foundation: logistic regression, Ridge, Lasso, and neural networks build on it. It is interpretable: each weight says directly how much the prediction changes if that feature increases by one unit. But before interpreting weights—you must understand what we are minimizing.