Graphics Reference
In-Depth Information
Table 4.4
Average ranks for the Rule Induction Learning methods
C45 Ripper
PART
Slipper AQ CN2 SRI Ritio Rules-6 Avg.
Ranks
IM
5
8.5
1
4
6.5
10
6.5
6
5
5.83
4
EC
2.5
8.5
6.5
1
6.5
5.5
6.5
6
1
4.89
3
KNNI
9
2.5
6.5
11
11
5.5
11.5
11
11
8.78
11
WKNNI
11
2.5
6.5
7
6.5
1
11.5
6
11
7.00
8
KMI
5
2.5
6.5
3
6.5
5.5
9.5
12
7.5
6.44
6
FKMI
7.5
2.5
6.5
10
2
5.5
1
2
3
4.44
1
SVMI
1
5.5
6.5
7
1
5.5
6.5
6
2
4.56
2
EM
13
12
6.5
7
12
13
3
6
4
8.50
10
SVDI
11
11
6.5
12
10
12
9.5
10
11
10.33
12
BPCA
14
13
13
7
13
14
13
13
13
12.56
14
LLSI
11
5.5
6.5
7
6.5
11
3
6
7.5
7.11
9
MC
7.5
8.5
6.5
2
6.5
5.5
3
6
7.5
5.89
5
CMC
5
8.5
12
13
3
5.5
6.5
1
7.5
6.89
7
DNI
2.5
14
14
14
14
5.5
14
14
14
11.78
13
Table 4.5
Average ranks for the Black Box methods
RBFN RBFND RBFNI LOG LVQ MLP NB
ν
-SVM C-SVM SMO Avg. Ranks
IM
9
6.5
4.5
6
3.5
13
12 10
5.5
5.5
7.55 10
EC
1
1
1
3
7
8.5
10 13
1
2
4.75
1
KNNI
5
6.5
10.5
9
7
11
6.5 8
5.5
5.5
7.45
9
WKNNI 13
6.5
4.5
10
10
4.5
6.5 4.5
5.5
5.5
7.05
6
KMI
3.5
2
7
3
11
3
4.5 8
5.5
9
5.65
2
FKMI
12
6.5
10.5
3
1.5
4.5
11 4.5
5.5
3
6.20
3
SVMI
2
11.5
2.5
7.5
3.5
1.5
13 8
11
9
6.95
5
EM
3.5
6.5
13
12
12.5 10
4.5 4.5
10
11.5
8.80 11
SVDI
9
6.5
7
11
12.5 8.5
3
11.5
12
11.5
9.25 12
BPCA 14
14
14
13
7
14
2
2
13
13
10.60 14
LLSI
6
6.5
10.5
7.5
7
6.5
9
4.5
5.5
9
7.20
7
MC
9
6.5
10.5
3
7
6.5
8
11.5
5.5
5.5
7.30
8
CMC
9
13
2.5
3
1.5
1.5
14 14
5.5
1
6.50
4
DNI
9
11.5
7
14
14
12
1
1
14
14
9.75 13
imputation methods and we cannot choose the best method from among these three.
On the other hand, BPCA and DNI are the worst methods.
In Table
4.5
we can observe the rankings associated with the methods belonging to
the black-boxesmodeling category. As can be appreciated, for black-boxesmodelling
the differences between imputation methods are even more evident. We can select
the EC method as the best solution, as it has a difference of ranking of almost 1 with
KMI, which stands as the second best. This difference increases when considering
