Training data is the collection of examples used to teach a machine learning model patterns, relationships, or desired behavior.
Its quality, representativeness, labeling, governance, and provenance strongly affect model performance, bias, and reliability.