On the Search Engine Result Clickstreams datasets MIPSVM is between 1
and 1.5 orders of magnitude faster than C4.5, Naive Bayes and Logistic Regres-
sion. Compared to Voted Perceptron it is about 2.5 orders of magnitude faster
(figure 11).
1
10
100
1000
10000
Spider or Person
R
e
l
a
t
i
v
e
r
u
n
t
i
m
e
(
l
o
g
a
r
i
t
h
m
i
c
s
c
a
l
e
)
MIPSVM
C4.5
Logistic
Voted Perceptron
Naive Bayes
Fig. 12. Performance - person vs agent clickstreams
On the Spider and Person Clickstreams dataset MIPSVM is approximately
1 order of magnitude faster than Naive Bayes, 2 orders of magnitude faster than
C4.5, 1.5 orders of magnitude faster than Logistic Regression and 3 orders of
magnitudes faster than Voted Perceptron (figure 12.
3
Conclusion
Finally MIPSVM was been compared to other classifiers on three main types of
classification data:
1. Classification datasets from the UCI Machine Learning Repository
2. Classicication datasets from the Zereal Massively Multiplayer Online Game
Simulator
3. Classification datasets from a Web Usage Log extracted from www.jfipa.org
3.1
Computational Performance
MIPSVM is faster than the other classifiers in all experiments. This can be
caused by several reasons: the algorithms, implementation overhead, program-
ming language and optimizations in compiler and virtual machines, external
libraries used (the ATLAS linear algebra library is used in the qimplementation
of MIPSVM [9]).
146
Empirical Comparison of MIPSVM with existing classifiers