What is feature leakage and how is it different from data leakage?

Feature leakage occurs when input features contain information about the target, while data leakage is any unintended information flow from outside training data.

Feature leakage is a specific type of data leakage where a feature directly or indirectly reveals the target variable (e.g., using future data or post-outcome variables). Data leakage is broader and includes any contamination between training and testing data. Feature leakage is especially dangerous because it leads to unrealistically high performance during training and evaluation.