Why does feature scaling matter for some models more than others?

Models based on distances, regularized coefficients, or gradient-based optimization can be strongly affected by the relative scale of features. A feature measured in thousands can dominate one measured in fractions even if it is less informative. Tree-based splits generally do not require standard scaling of numeric features in the same way. Fit scaling on training data only and include it in the evaluation pipeline. Inspect outliers before choosing standardization or a robust alte ative. Scaling does not repair leakage, invalid units, or a fundamentally unsuitable feature.