Spot the data leakage

Machine LearningIntermediatemultiple choiceSample content

An analyst standardises all features using the mean and SD of the full dataset, then splits into train and test sets. What is the problem?

None — scaling before splitting is standard.Information from the test set leaks into training, so test performance is optimistic.Standardisation always reduces accuracy.The model cannot be trained on scaled data.