What is 'Data Leakage' in Machine Learning?

What is 'Data Leakage' in Machine Learning?

Approach: Explain using information from outside the training dataset to create the model.

Occurs when information from the target variable or the test set 'leaks' into the training process, leading to over-optimistic performance metrics that cannot be replicated in production.

Follow-up: Give an example (e.g., using future data for past predictions).

Key concepts: target leakage, train-test contamination, over-optimistic.