Definition
The inadvertent creation of a statistical dependency between the training and testing datasets, often resulting in overly optimistic model performance estimates.
In short: The inadvertent creation of a statistical dependency between the training and testing datasets, often resulting in overly optimistic model performance estimates.