A new concept of probability metric and its applications in approximation of scattered data sets