Information Technology Reference
In-Depth Information
6.4.2
Constructing Rule Evaluation Models on Artificial Evaluation
Labels
We also evaluated our rule evaluation model construction method using rule sets
obtained from five datasets of the UCI machine learning repository to confirm
the lower limit performances on probabilistic class distributions.
We selected the following five datasets: anneal, audiology, autos, balance-
scale, breast-cancer, breast-w, colic, and credit-a. With these datasets, we ob-
tained rule sets with bagged PART, which repeatedly executes PART [35] on
the bootstrapped training datasets.
For these rule sets, we calculated 39 objective indices as attributes of each
rule. With regard to the classes of these datasets, we used three class distribu-
tions with multinomial distributions. Table 6.4 shows us a process flow diagram
for obtaining these datasets and their descriptions, with three different class
Table 6.4. The datasets of the rule sets learned from the UCI benchmark datasets
#Mined
Rules
#Class labels
%Def. class
L1 L2 L3
(0.30) (0.35) (0.35)
Distribution I
anneal
95
33
39
23
41.1
audiology
149
44
58
47
38.9
autos
141
30
48
63
44.7
balance-scale
281
76
102 103
36.7
breast-
122
41
34
47
38.5
breast-w
79
29
26
24
36.7
colic
61
19
18
24
39.3
credit-a
230
78
73
79
34.3
Distribution II
(0.30) (0.50) (0.20)
anneal
95
26
47
22
49.5
audiology
149
44
69
36
46.3
autos
141
40
72
29
51.1
balance-scale
281
76
140
65
49.8
breast-
122
40
62
20
50.8
breast-w
79
29
36
14
45.6
colic
61
19
35
7
57.4
credit-a
230
78
110
42
47.8
Distribution III
(0.30) (0.65) (0.05)
anneal
95
26
63
6
66.3
audiology
149
49
91
9
61.1
autos
141
41
95
5
67.4
balance-scale
281
90
178
13
63.3
breast-
122
42
78
2
63.9
breast-w
79
22
55
2
69.6
colic
61
22
36
3
59.0
credit-a
230
69
150
11
65.2
 
Search WWH ::




Custom Search