Information Technology Reference
In-Depth Information
In this case, a new sample
x
was considered to match (satisfy) a rule
R
, denoted by
x
R
if the hypersphere with center at
x
and radius
v
intercepts the hyperrectangle
defi ned by
R
. h e parameter
α
denotes a coe
cient of sensitivity, which for a specifi c
rule determines the trade-off between the volume covered by it and its interception
with the self-set. h us, the total fi tness is calculated as the sum of the fi tness of all
evolved rules minus the overlapping between the hyperrectangles defi ned by the
rules.
A “sequential niching” algorithm was used to evolve a set of suitable rules to
cover the nonself space (Dasgupta and Gonzalez, 2002). In this case, several runs
of the evolutionary algorithm were needed, and a suitable rule was obtained at the
end of each run. Gonzalez and Dasgupta (2002) used a “deterministic crowding
(DC) niching technique,” which performed better in covering the complementary
space; thus, a smaller set of rules was obtained to estimate the amount of devia-
tion more precisely. h is new niching technique also allowed the generation of
multiple rules in a single run instead of multiple runs as in “sequential niching.” In
the DC niching approach, a distance measure between two detectors
c
and
p
was
defi ned as
∈
∩
volume p
()
volume p
(
c
)
dist c p
(, )
volume p
()
where
c
and
p
are child and parent individuals, respectively. Note that this distance
measure is not symmetric because it gives more importance to the area of the parent
that is not covered.
h e pseudocode for negative selection with detection rules (NSDR) algorithm
using the DC are given in the following.
NS Algorithm 5: NS-detector-rules (
S', num
_
levels,
{
v
1
, …,
v
numLevels
})
Input:
S
: set of self samples
num_levels:
number of deviation levels
{
v
1
, …,
v
numLevels
}: allowed variability for each level
Output:
Change Detection rules.
=
1
to
num_levels
2 initialize population with random individuals
3
For
j
1
For
i
=
1
to
num_gen
=
4
For
k
1 to pop_size/2
5
select two individuals, (
parent
1,
parent
2), with uniform prob-
ability and without replacement
Search WWH ::
Custom Search