Information Technology Reference
In-Depth Information
of
Table 1
shows the result. But even in this case the classification e
%
ciency and
the number of attractors required to represent the data set doesn't deteriorate
much.
Table 2.
Clusters Detection by MACA based Classifier,
n
= 100,
D
min
= 20,
d
max
=5
Combi
n
Value Performance (%)
of Clusters of
m
Training Testing
A
&
B
,
C
&
D
2
95.90
92.30
4
99.82
97.10
A
&
C
,
B
&
D
2
94.50
92.30
4
98.70
96.62
A
&
D
,
B
&
C
2
94.60
90.40
4
99.20
96.82
5.4
Expt 4a: Experiment with Data Sets Having Implicit Clusters
Even if the data set between two classes have high overlap (
d − d
), the classifier
functions well if the classes have implicit clusters among them. The following
experiment is performed to validate this observation.
To perform this experiment, we first randomly generate four pivot patterns
(
Fig.11
) which are separated by a fixed hamming distance (say
D
min
). Then
around each pivot, we randomly generate a cluster of
p
number of patterns within
d
max
distance (maximum hamming distance between a pivot and an element in
the associated cluster); where
d
max
<
=
D
min
/
2. It is seen that
MACA
based
classifier perform classification task very well even though the distribution of
patterns of each class is not ideal. The
Table 2
gives the detail result of cluster
detection.
The
Column III
and
IV
depict the classification e
%
ciency of the training and
testing data set respectively at different values of
m
. In the two attractor basins
(say
a
&
b
) which
Cluster A
&
B
occupy, it is observed that 99% of patterns of
Cluster A
occupies attractor
a
, while 99% of patterns of
Cluster B
occupies the
attractor
b
. It is similar in case of the class formed from
C
and
D
. This is in line
with the theoretical formulation that each attractor basin generates a cluster of
patterns having lesser hamming distance.
6 Conclusion
The paper presents the detailed design and analysis of Cellular Automata based
Pattern Classification technique. Theoretical formulation and experimental re-
sults prove beyond doubt the scope of the machine becomes really popular in
industry in the near future.
