Modified Mahalanobis Taguchi System for Imbalance Data Classification
Table 2
Summary of the dataset used in the study.
Number
Dataset
Class
# variables
Number of observations
-ratio
IR ratio
value
Statistically
Major/minor
Negative
Positive
Significant
(1)
Abalone
Remainder/Class 24
8
4175
2
7.797
2088 : 1
0.0000
Yes
(2)
Abalone
Remainder/Class 22
8
4171
6
0.814
695 : 1
0.0000
Yes
(3)
Abalone
Remainder/Class 23
8
4168
9
0.661
463 : 1
0.0000
Yes
(4)
Abalone
Remainder/Class 3
8
4162
10
8.227
417 : 1
0.0028
Yes
(5)
Abalone
Remainder/Class 21
8
4165
12
1.244
347 : 1
0.0000
Yes
(6)
Abalone
Remainder/Class 21
8
4163
14
1.000
297 : 1
0.0000
Yes
(7)
Abalone
Remainder/Class 21
8
4151
22
1.019
189 : 1
0.0000
Yes
(8)
Abalone
Remainder/Class 21
8
4151
26
0.868
160 : 1
0.0000
Yes
(9)
Abalone
Remainder/Class 19
8
4145
32
0.555
130 : 1
0.0000
Yes
(10)
ECOLI
Remainder/Class OML
7
331
5
56.509
66 : 1
0.0000
Yes
(11)
Welding
Normal/Expulsion
28
316
6
18.837
53 : 1
0.0122
Yes
(12)
Yeast
Remainder/Class ME2
8
1433
51
1.144
28 : 1
0.0000
Yes
(13)
Shuttle
Remainder/Class 5
9
41042
2458
11.513
17 : 1
0.0000
Yes
(14)
Glass
Remainder/Class 7
9
185
29
2.806
6 : 1
0.8156
No
(15)
Heart disease
Absence/Presence
13
150
120
0.872
1.25 : 1
0.0000
Yes
Fisher discriminant ratio; data overlapping index, imbalance ratio = ; based on Kruskal-Wallis nonparametric test; is there any statistical significant difference among classifiers performance (yes/no)? [40].