Databases Reference
In-Depth Information
from each class proportional to its size, except that the smallest class is
completely included. This data set is again divided into training data with
5092 records and testing data with 6890 records. All the intrusion detection
models are trained and tested with the same set of data. As the data set has
five different classes we perform a five-class classification. The normal data
belongs to class one, probe belongs to class two, denial of service (DoS)
belongs to class three, user to root (U2R) belongs to class four and remote
to local (R2U) belongs to class five.
HAPSO parameters have been choosen without performing a
preliminary tuning phase, rather they have been set on the basis of
the experience gained of multivariate optimization problem. Namely their
values are the following:
Swarm size, N
50
c
1
2.0
c
2
1.8
Inertia weight,
w
min
0.4
Inertia weight,
w
max
0.9
5.5.3.
Results
The accuracy of the classifier is measured through the confusion matrix.
As it is a stochastic algorithm we reported the average training set results
obtained on ten simulations in Table 5.4. The top-left entry of the following
matrix shows that 1394 of the actual normal test set were detected to
be normal; the last column indicates that 99.57% of the actual normal
data points were detected correctly. In the same way, for Probe 658 of the
actual attack test set were correctly detected; the last column indicates
that 94.00% of the actual Probe data points were detected correctly.
Similarly for Dos 98.22% of the actual DoS test data points were detected
correctly. 64.00% of the U2Su test data points are correctly classified.
In the case of R2U class 95.71% of the test data points are correctly
classified. The bottom row shows that 96.87% of the test set said to
be normal indeed were normal and 89.52% of the test set classified, as
probe indeed belongs to probe. The overall accuracy of the classification
is 97.69%.
The most evident aspect of the results on this dataset is the feature
selection capability demonstrated by the HAPSO/Bayesian classifier. The
