Information Technology Reference
In-Depth Information
three values of
m
noted in
Column II
of
Table 1
. The classification performance
improves at a very slow gradient on further increase of
m
.
dmax
Dmin
Class I
A
B
Class II
D
C
Fig. 11.
Clusters Detection by two-class classifier
Table 1.
Experiment to find out the value of
m
Size Value Curve a-a' Curve b-b' Curve c-c'
(
n
) of
m
Training Testing Training Testing Training Testing
20
2
85.40
85.60
83.20
82.00
81.20
72.40
3
96.10
94.35
92.20
93.35
92.20
83.35
40
3
98.20
97.75
97.60
96.80
77.60
66.80
4
98.35
98.85
97.20
97.45
94.28
87.45
60
3
98.55
97.75
96.90
96.05
86.98
77.55
4
98.50
98.00
96.90
96.05
91.90
86.60
80
3
98.81
98.65
98.70
97.70
81.70
76.35
4
99.15
99.20
98.70
97.70
88.79
87.30
5
99.75
99.70
98.30
97.30
91.65
87.10
100
3
99.65
99.25
98.30
97.45
86.40
77.95
4
99.67
99.35
98.40
97.30
83.10
80.35
5.2
Expt 2: Experiment Done on
a
−
a
and
b
−
b
It shows that classification accuracy is above 98% in most of the cases. Moreover,
the prediction accuracy is also almost equal to the classification accuracy. This
validates the theoretical foundation that
MACA
basins are natural clusters.
5.3
Expt 3: Experiment Done on Data Set
c
−
c
With increasing overlap of data set (curve
c − c
), a significant gap is created
between classification e
%
ciency of test and trained data. The
Column V
&
VI