IEEE Computational Intelligence Magazine - November 2022 - 64

TABLE V The AveAcc values of five different algorithms on 10 datasets.
DATASET
EMT-MC
Balance
Contraceptive
Iris
Newthyroid
Wine-quality-red
Dermatology
Poker1
Segment
Led7digit
Penbased
win/tie/loss
0.8987 0.0004
0.5147 0.0010
0.9595 0.0012
0.9708 0.0010
0.3643 0.0022
0.9390 0.0006
0.6727 0.0002
0.8485 0.0009
0.7107 0.0017
0.8958 0.0001
RUS
0.6789 0.0011(+)
0.4898 0.0003(+)
0.8493 0.0022(+)
0.9022 0.0024(+)
0.3475 0.0088(+)
0.9044 0.0026(+)
0.5184 0.0013(+)
0.7143 0.0005(+)
0.4197 0.0056(+)
0.7900 0.0001(+)
10/0/0
ELM
MC2ESVM
0.8836 0.0076(+)
0.5073	0:0002ð)
0.7822 0.0042(+)
0.8605 0.0038(+)
0.5046 0.0003(-)
0.8514 0.0012(+)
0.6365 0.0001(+)
0.7976 0.0003(+)
0.6510 0.0034(+)
0.6296 0.0001(+)
8/1/1
0.6715 0.0005(+)
0.5096	0:0003ð)
0.7259 0.0001(+)
0.9129 0.0002(+)
0.2970 0.0001(+)
0.9141 0.0003(+)
0.4702 0.0001(+)
0.5685 0.0001(+)
0.6783 0.0003(+)
0.7128 0.0001(+)
9/1/0
MROCF
0.8713 0.0022(+)
0.5021 0.0005(+)
0.9239 0.0013(+)
0.9578 0.0012(+)
0.3271 0.0020(+)
0.9193 0.0023(+)
0.5216 0.0016(+)
0.8387	0:0005ð)
0.7154 0.0019()
0.8808 0.0020(+)
8/2/0
TABLE VI The MAUC values of five different algorithms on 10 datasets.
DATASET
EMT-MC
Balance
Contraceptive
Iris
Newthyroid
Wine-quality-red
Dermatology
Poker1
Segment
Led7digit
Penbased
win/tie/loss
nine datasets. The better performance of
EMT-MC mainly lies in the fact that
unlike MROCF that handles each
binary classification sub-problem separately,
in EMT-MC, " ill-solved " classifiers
are aided by the " well-solved " classifiers,
which helps the final muticlass classifier
Balance
Contraceptive
Iris
Newthyroid
Wine-quality-red
Dermatology
Poker1
Segment
Led7digit
Penbased
EMT-MC
87
145
21
34
1306
648
2729
3075
4796
8801
RUS
0.9606 0.0001
0.7000 0.0010
0.9959 0.0001
0.9995 0.0001
0.8043 0.0007
0.9943 0.0001
0.8127 0.0006
0.9801 0.0001
0.9644 0.0001
0.9914 0.0001
RUS
0.7988 0.0003(+)
0.6875	0:0004ð)
0.9184 0.0006(+)
0.9664 0.0004(+)
0.7385 0.0007(+)
0.9934	0:0001ð)
0.6098 0.0008(+)
0.9614 0.0001(+)
0.9148 0.0001(+)
0.9883	0:0001ð)
7/3/0
ELM
MC2ESVM
0.7542 0.0008(+)
0.6968	0:0007ð)
0.9381 0.0007(+)
0.9476 0.0007(+)
0.7656 0.0005(+)
0.9969 0.0001()
0.6781 0.0013(+)
0.9699 0.0001(+)
0.9523 0.0001(+)
0.9870	0:0001ð)
7/3/0
with higher quality. Based on the
statistics in Tables V and VI, it can be concluded
that the proposed algorithm is a
competitive method to obtain the multiclass
classifier with good quality.
In addition, the time comparison
of different multiclass classification
TABLE VII The time comparison (s) of five different algorithms on 10 datasets.
DATASET
ELM
18
MC2ESVM
0.2 73.4
0.3
14
0.09
0.08
0.6
0.11
1.7
0.11
0.13
5
0.9
1.6
34
6
116
51
14
243
0.4
0.4
2.3
0.3
247
14
0.3
191
0.7950 0.0005(+)
0.6658 0.0008(+)
0.8516 0.0001(+)
0.9681 0.0001(+)
0.6909 0.0022(+)
0.9835 0.0009(+)
0.7131 0.0026(+)
0.8875 0.0001(+)
0.9431 0.0001(+)
0.9372 0.0001(+)
10/0/0
MROCF
0.9530	0:0003ð)
0.7095 0.0002()
0.9786 0.0001(+)
0.9767 0.0001(+)
0.6924 0.0052(+)
0.9894	0:0001ð)
0.7266 0.0042(+)
0.9646 0.0001(+)
0.9583	0:0001ð)
0.9832 0.0001(+)
6/4/0
MROCF
73
85
64
68
429
322
654
545
1189
2147
algorithms on 10 datasets is shown in
Table VII. It can be observed from
the table that RUS has the fastest running
time, since it uses sampling technique
without decomposing the
original multiclass classification problem
into many binary sub-problems.
All of the other four algorithms
(ELM, MC2ESVM, MROCF and
EMT-MC) adopted decomposition
scheme, thus, their time is longer than
RUS. By taking a closer look at four
decomposition-based algorithms, it
can be found that ELM and
MC2ESVM have much less running
time than MROCF and the proposed
EMT-MC. The reason mainly lies in
thefactthatELM and MC2ESVM
adopt OVA decomposition scheme
which only needs to solve m subproblems,
while MROCF and EMT64
IEEE COMPUTATIONAL INTELLIGENCE MAGAZINE | NOVEMBER 2022
IEEE Computational Intelligence Magazine - November 2022

Table of Contents for the Digital Edition of IEEE Computational Intelligence Magazine - November 2022
