Data Leakage

Data leakage occurs when information from outside the training dataset is inadvertently used to create the model, giving an overly optimistic estimate of its performance. This often happens if data that would not be available in a real-world scenario is included during training, or if future information 'leaks' into past observations.