Graphics Reference
In-Depth Information
Table 5.8 Test accuracy, RLA results and p-values on data sets with class noise. Cases where the
baseline classifiers obtain more ranks than the OVO version in the Wilcoxon's test are indicated
with a star (*)
Uniform random class noise Pairwise class noise
C4.5 Ripper 5-NN C4.5 Ripper 5-NN
Base OVO Base OVO Base OVO Base OVO Base OVO Base OVO
0% 81.66 82.7 77.92 82.15 82.1 83.45 81.66 82.7 77.92 82.15 82.1 83.45
10% 80.5 81.71 71.3 79.86 81.01 82.56 80.94 81.86 75.94 80.71 81.42 82.82
20% 78.13 80.27 66.71 77.35 79.55 81.36 79.82 81.03 74.77 79.62 79.41 81.01
30% 75.22 78.87 62.91 74.98 77.21 79.82 78.49 79.26 73.38 78.05 75.29 76.81
40% 71.1 76.88 58.32 72.12 73.82 76.83 76.17 76.91 71.6 76.19 69.89 71.65
50% 64.18 73.71 53.79 67.56 68.04 72.73 63.63 63.52 67.11 65.78 64.02 65.52
0%
-
-
-
0.007
0.0002
0.093
10% 0.0124
0.0001
0.0036
0.0033
0.0003
0.0137
20% 0.0028
0.0001
0.0017
0.0017
0.0002
0.0022
30% 0.0002
0.0001
0.0013
0.009
0.0001
0.01
40% 0.0002
0.0001
0.0111
0.0276
0.0003
0.004
50% 0.0001
0.0001
0.0008
0.5016(*)
0.0930(*)
0.0057
0% - - - - - -
10% 1.56 1.28 9.35 2.79 1.46 1.12 0.91 1.01 2.38 1.72 0.89 0.82
20% 4.63 3.15 15.44 5.86 3.48 2.67 2.39 2.13 3.52 3.03 3.29 2.93
30% 8.36 4.9 20.74 8.82 6.4 4.53 4.16 4.49 5.13 4.84 8.05 7.81
40% 13.46 7.41 26.72 12.35 10.3 8.11 7.01 7.38 7.25 7.12 14.37 13.68
50% 21.87 11.29 32.74 18.1 17.47 13.12 21.03 22.28 12.22 18.75 21.06 20.67
0%
-
-
-
-
-
-
10% 0.5257
0.0001
0.1354
0.8721(*)
0.3317
0.3507
20% 0.0304
0.0001
0.0479
0.0859
0.4781
0.0674
30% 0.0006
0.0001
0.0124
0.6813
0.6542
0.3507
40% 0.0001
0.0001
0.0333
0.6274
0.6274(*)
0.0793
50% 0.0001
0.0001
0.0015
0.0400(*)
0.0001(*)
0.062
non-OVO version from the Wilcoxon's test. The few exceptions where the baseline
classifiers obtainmore ranks than theOVOversion in theWilcoxon's test are indicated
with a star next to the p-value.
For random class noise the test accuracy of the methods using OVO is higher in
all the noise levels. Moreover, the low p-values show that this advantage in favor of
OVO is significant. The RLA values of the methods using OVO are lower than those
 
Search WWH ::




Custom Search