Should I scale features before or after splitting the dataset?

Split first. Fit the scaler on training data only, then transform validation and test data with that fitted scaler. Fitting on the full dataset leaks evaluation information.

Reference: Official documentation.

AI-assisted DevCircle editorial answer. Corrections welcome.