Databases Reference
In-Depth Information
Table 3.1.
Description of datasets.
Dataset
Patterns
Attributes
Classes
Patterns
Patterns
Patterns
in class 1
in class 2
in class 3
AUSTRA
690
14
2
307
383
—
BALANCE
625
4
3
288
49
288
CREDIT(GERMAN)
1000
20
2
700
300
—
HABERMAN
306
3
2
225
81
—
HOUSE
435
16
2
267
168
—
IRIS
150
4
3
50
50
50
MONK
124
6
2
62
62
—
PIMA
768
8
2
268
500
—
WBC
699
9
2
458
241
—
WINE
178
13
3
59
71
48
results of ten different datasets collected from UCI Repository of Machine
Learning database.
104
A brief description of the properties of these dataset
is presented in Table 3.1.
For these simulations, we have preferred the two-fold cross validation.
Each database is divided into two sets. For the division, the records are
segregated depending on their class labels. Randomly we pick up the records
belonging to a class and place it in one of the sets. The distribution of
patterns to different sets of the databases is summarized in Table 3.2.
We have used PSO technique to evolve the architecture of the ANN for
the purpose of classification of the databases presented in Table 3.2. While
one set is used to evolve architecture, the other set is used for the purpose of
validation. Alternately both the sets are used to evolve the architecture and
for validation. The parameters of PSO used for the purpose of simulation
are presented in Table 3.3.
We have used ten real life databases for the purpose of evolving ANN
with PSO. The classification accuracies obtained from the training set and
testing set are presented in Table 3.4. The results obtained from the training
sets and test sets of a data base are averaged and the comparison of
the performance is presentedinFig.3.4.The
x
-axis values 1
,
2
,
3
,...,
10
shown in Fig. 3.4 represents the databases AUSTRA, BALANCE,
CREDIT, HABERMAN, HOUSE, IRIS, MONK, PIMA, WBC, WINE
respectively.
Further, for the purpose of evolution of the ANN architecture, we have
restricted the number of hidden layers maximum to three and the number of
hidden neurons in each hidden layer to ten. The details of the architecture
