Databases Reference
In-Depth Information
concern, we may use a certain number of rules, say
K
, whose antecedents are
the closest to the new feature vector
F
. Then, only
K
BPAs would need to
be fused using the DRC.
To ensure the integrity of the generated rule set, the classi
fier deve
loped
as abo
ve was used to classify the feature vectors
F
(
k
)
,
= 1
,N
R
(
k
)
,k
=
1
,N
TC
, of the training data set itself. The classification error associated with
F
(
k
)
can be considered to reveal that the rule set does not possess su
cient
information to correctly identify the training data set. With this observation
in mind,
R
Pruned
was supplemented by each training data instance whose
feature vector
F
(
k
)
{
}
was not correctly classified; the confidence measure of such
a rule was allocated a value of 1
.
0. This refined rule set is what constitutes
our proposed ARM-KNN-BF classifier.
2.8 An Example of Rule Generation
In this section, we use a slightly modified variation of a data set from [22] to
clarify and illustrate the various steps involved in our proposed rule generation
algorithm. The data set being considered possesses three features and two
classes. See Table 1. The FoDs corresponding to the features are as follows:
Outlook :
Θ
f
1
=
{
sunny
,
overcast
,
rainy
}
;
Humidity :
Θ
f
2
=
{
LOW
,
MEDIUM
,
HIGH
}
;
Windy :
Θ
f
3
=
{
TRUE
,
FALSE
}
;
Decision :
Θ
C
=
{
Play
,
Don't Play
}
.
(24)
Tabl e 1 .
The training data set under consideration
Outlook
Humidity
Windy
Decision
sunny
MEDIUM
TRUE
Play
sunny
LOW
FALSE
Play
sunny
MEDIUM
FALSE
Play
overcast
HIGH
FALSE
Don't play
rainy
MEDIUM
FALSE
Don't play
rainy
HIGH
TRUE
Θ
C
overcast
LOW
FALSE
Play
sunny
MEDIUM
TRUE
Play
rainy
HIGH
FALSE
Don't play
overcast
MEDIUM
FALSE
Don't play
overcast
HIGH
TRUE
Θ
C
sunny
HIGH
TRUE
Θ
C
sunny
LOW
TRUE
Play
overcast
LOW
TRUE
Play
sunny
MEDIUM
FALSE
Play
overcast
HIGH
TRUE
Don't play
sunny
LOW
TRUE
Play
