The specifications of the data sets are listed inTable 2. As in the real applications, the distributions ofthese data set are unknown and most of them are notnoisy-free. For each case, the training data set and testingdata set are randomly generated from its whole data setbefore each trial of simulation.