The curve is an error profile and traces the relationship between classification errors and tree size. It is called a Relative Cost curve or Relative Error and it is scaled between 0 and 1. Zero means no errors or a perfect fit and 1 represents the performance of random guessing.