B.5
Paired T-Tests for Classification Accuracy
157
B.5
Paired T-Tests for Classification Accuracy
This section shows the Minitab analysis output based on paired T-tests of classi-
fication accuracy of the proposed MIPSVM classifier implemented in the Incridge
tool with the C4.5, Logistic Regression, Naive Bayes and Voted Perceptron clas-
sifiers implemented in the Weka tool. Paired T-test was chosen since it allows
a "global summary" over all the classification experiments performed (uses the
difference between classifier results per row)
B.5.1
Classification Data C4.5 and Naive Bayes
Dataset/Classifier
MIPSVM
C4.5
Naive Bayes
P0
46.5116 46.5856 26.678
P1
84.4444 83.6957 53.804
P2
98
97.5904 95.181
P3
15.7143 23.1293 9.524
P4
47.0588 44.6927 6.145
P5
28.8889 30
5.556
P6
11
13.8614 17.822
P7
29.4737 33.5079 19.372
Intrasite Clickstr. 21.9172 45.0676 4.257
From Google or not
58.8485 58.6645 52.088
From SE or not
63.6768 63.5667 50.212
Spider or Person
69.5661 85.6264 70.92
Waveform
85.98
75.52
79.96
Segment
84.72
97.14
80.3
Letter
50.42
87.76
64.23
Players vs NPCs
98.125
98.125
100
Player Types
85.625
77.5
75
Players and NPCs
75.625
81.25
81.875
B.5.2
MIPSVM and C4.5 Comparison
The P-Value based on the T-test is 0.088, this is not significant according to
5% significance level, hence the experiments performed does not show significant
difference between MIPSVM and C4.5.
Welcome to Minitab, press F1 for help.
n
Paired T-Test and CI: MIPSVM; C4.5