“…Based on this, the outputs of testing data of VPMCD multi-classifier as well as their identifying rate are given in Table 5. Comparing with Table 4, it is easy to find that the X 1 -X 10 1 1(10) 1(10) 1(10) 1(10) 1(10) 1(10) X 11 -X 20 2 2(10) 2(10) 2(10) 2(10) 2(10) 2(10) X 21 -X 30 3 3(6), 7(3), 4(1) 3(10) 3(10) 3(0), 4(10) 3(10) 3(7), 4(3) X 31 -X 40 4 4(10) 4(10) 4(10) 4(10) 4(10) 4(10) X 41 -X 50 5 5(10) 5(10) 5(10) 5(10) 5(10) 5(10) X 51 -X 60 6 6(10) 6(10) 6(9), 7(1) 6(5), 7(5) 6(9), 7(1) 6(9), 7(1) X 61 -X 70 7 7(9), 6(1) 7(8), 6(1), 2(1) 7(10) 7(9), 6(1) 7 X 1 -X 10 1 1(10) 1(10) 1(10) 1(10) 1(10) 1(10) X 11 -X 20 2 2(10) 2(10) 2(10) 2(10) 2(10) 2(10) X 21 -X 30 3 3(10) 3(10) 3(10) 3(10) 3(10) 3(10) X 31 -X 40 4 4(10) 4(10) 4(10) 4(10) 4(10) 4(10) X 41 -X 50 5 5(10) 5(10) 5(10) 5(10) 5(10) 5(10) X 51 -X 60 6 6(10) 6(10) 6(10) 6(10) 6(8), 7(2) 6(9), 7(1) X 61 -X 70 7 7(8), 6(2) a 7 ( a r c h i v e s o f c i v i l a n d m e c h a n i c a l e n g i n e e r i n g 1 6 ( 2 0 1 6 ) 7 8 4 -7 9 4 identification rates of testing data without sorting by LS are all lower than that of features optimized by LS for all J (from 2 to 7). It indicates that the feature selection by using LS is essential and dominant.…”