They should give the same result because tree-based models are scale invariant.
Card 1 of 276. Answer: They should give the same result because tree-based models are scale invariant.
About · Quizzes · History of AI · To suggest new questions or report an error send me a dm.
Neither: they should give the same results. A decision tree splits on rules like feature ≤ threshold, and the best split depends only on how the rows are ordered by that feature, not on its units. Standardization and min-max scaling preserve that order for each feature, so they produce the same tree.
Example: incomes of 20,000, 35,000, 50,000 and 90,000, with a split at income ≤ 42,500 separating the first two rows. Measured in thousands, the split becomes income ≤ 42.5, with the same rows on each side and the same impurity reduction. Every candidate split maps one to one, so random forests and gradient boosting are unaffected too.
The precise property is invariance to monotonic transformations of each feature, with one edge case. A log transform keeps the order too, but a threshold halfway between log(35,000) and log(50,000) corresponds to about 41,833, so a new income of 42,000 is routed differently. Otherwise scaling matters only for other models in the pipeline, such as KNN or neural networks.
