Validating a machine learning model