Training data where some examples have incorrect labels, which can degrade model performance if not handled carefully.