Consolidated Trees: An Analysis of Structural Convergence - Data Mining: Theory, Methodology, Techniques, and Applications - page 48

Table 3. Results of % Common for every domain. C4.5 100 , CTC and C4.5 union .

%Common C4.5 100 CTC C4.5 union

5 10 20 30 40 50 5 10 20 30 40 50

Breast-W 60 87 95 97 98 99 99 73 78 86 89 89 89

Heart-C 13 26 36 47 52 55 64 14 19 28 34 36 39

Hypo 56 70 74 80 89 94 92 42 51 52 57 57 57

Lymph 30 59 72 80 86 84 91 59 67 76 76 78 80

Credit-G 8 15 20 29 32 37 44 12 16 22 26 30 30

Segment210 19 25 31 37 39 35 45 26 33 41 45 56 47

Iris 69 82 84 87 90 87 92 64 70 77 74 76 81

Glass 15 22 25 27 28 26 28 24 31 41 44 43 46

Voting 53 72 84 91 92 90 92 51 58 67 69 72 70

Hepatitis 16 33 43 50 55 61 58 20 26 35 41 45 50

Soybean-L 31 46 53 64 70 69 72 52 60 66 69 72 74

Sick-E 46 53 56 59 63 65 67 18 17 18 18 22 21

Liver 7 9 13 17 21 24 27 6 9 17 19 23 24

Credit-A 23 30 36 48 53 58 55 20 24 30 35 38 39

Vehicle 14 15 17 20 21 23 23 18 21 24 25 29 32

Breast-Y 21 39 50 57 70 76 81 26 38 52 58 61 66

Heart-H 24 34 40 52 60 63 68 31 28 34 35 40 43

Segment2310 29 35 40 44 49 49 50 31 36 43 46 51 56

Spam 5 9 11 13 15 16 16 6 9 10 10 15 15

Faithful 18 29 30 31 36 39 31 7 7 9 10 11 11

Average

28 40 45 51 56 57 60 30 35 41 44 47 48

C4.5union
C4.5union(range)
CTC
CTC(range)
C4.5100

70

65

60

55

50

45

40

35

30

25

5 10 20 30 40 50

Number_Samples

Fig. 3. Averages of % Common for CTC, C4.5 100 and C4.5 union (5 times 10 folds, 20 databases)

For each database, average %Common values of the 5 runs and 10 folds are
calculated and every point in the graphic represents the average of the 20 databases. For
CTC and C4.5 union two studies are presented. In the first one, trees built with identical
values for Number_Samples parameter are compared (diamonds). In the second one,
called “range” in Fig. 3, trees with different values of Number_Samples are compared
(squares). In this case the point corresponding to N_S = 20 represents the %Common
value obtained from the comparison of every tree built with N_S ≥ 20. Being this value
(59) larger than %Common (51) means that there are trees built with 30, 40 or 50