Information Technology Reference
In-Depth Information
Table 4.6.
The confusion matrix of the
Lymphography
database using
PAT, C4.5 and
OAT
a b c d ¡- classified as
0000a=n mal
0 540b=metastases
01 00c=malign-lymph
0002d=fib s
a b c d ¡- classified as
0000a=n mal
0 540b=metastases
09 20c=malign-lymph
0002d=fib s
a b c d ¡-classified as
0000a=n mal
0 360b=metastases
07 40c=malign-lymph
0002d=fib s
Table 4.7.
Tests performed on
Mushroom
database
¬
well classif. 50%
Threshold well classif.
0.1
80.82%
19.17 %
PAT
% 0.2
75.34%
24.65 %
C4.5
58.90%
41.09%
OAT
67.12%
32.87%
Table 4.8.
The confusion matrix of the
Mushroom
database using
PAT, C4.5 and
OAT
a b ¡-classified as
33 5 a = e
13 22 b = p
a b ¡-classified as
33 5 a = e
25 10 b = p
a b ¡-classified as
35 3 a = e
21 14 b = p
2values
e, p
. The test data has 73 objects. The missing values rates are: 75%
for the attribute
odor
, 26% for
stalk-shape
, 52% for
stalk-root
, 23% for
veil-type
,
27% for
spore-print-color
and 19% for
ring-type
. We remark that in Tables 4.7,
4.8 our results are better than those given by C4.5 and
OAT
.
Finally,wepresentthetestsperformedonthe
Zoo
database [20]. The training
data has 101 instances and 17 attributes. The class takes 7 values:
mammal, bird,
reptile, fish, amphibian, insect, invertebrate
. The test data has 65 instances. The
missing values rates in the test data are: 21% for the attribute
feathers
, 29%
for
milk
, 16% for
airborne
, 16% for
aquatic
, 13% for
predator
and 13% for the
attribute
legs
. The result of testing is given in Table 4.9. From Tables 4.10, 4.11
we remark that when the class is
bird
,wehave17objectsinthetestdata,16
of which are misclassified with C4.5, 4 of which are misclassified with
OAT
and
only 1 object is misclassified with
PAT
.