A comparison of MCC and CEN error measures in multi-class prediction