Information Technology Reference
In-Depth Information
Algorithm 7.4
Selective accommodation of previous minority class examples
Inputs:
1: current timestamp
t
.
2: current training data chunk:
={
(
x
1
,y
1
),...,(
x
m
,y
m
)
}
3: current data set under evaluation:
S
(t)
x
1
,...,
x
n
}
T
(t)
={
4: minority class data queue:
Q
5: soft-typed base classifier:
L
6: selective accommodation method:
d
/*
d
could be
SERA
,
MuSeRA
,or
REA
*/
7: post-balanced ratio
f
/* desired minority class to majority class ratio. */
8: hypotheses set
H
={
h
1
,h
2
,...,h
t
−
1
}
Procedure:
9:
for
t
:1
→
...
do
10:
(t)
(t)
,
N
(t)
S
←{
P
}
(t)
(t)
/* Assume
||
P
|| =
p
and
||
N
|| =
q
*/
if
(
||
P
(t)
|| + ||
Q
||
)/
||
S
(t)
||
<
=
f
then
11:
←
L(
{
S
(t)
,
Q
}
)
12:
h
t
13:
else
14:
n
←
f
×||
S
t
|| − ||
P
t
||
15:
m
←{}
16:
for
x
j
∈
Q
do
17:
if
d
=
REA
then
(t)
)
/* calculates the
k
-nearest neighbors of
x
j
within
18:
K
←
k-nearest-neighbor
(
x
j
,
S
(t)
*/
S
m
,
k
∈
K
(t)
]]
19:
}
/* number of minority cases within the
k
-nearest neighbor of
x
j
*/
m
←{
[[
k
∈
P
20:
else
(x
j
−
μ)
T
−
1
(x
j
−
μ)
/*
μ
and
are the mean and the covariance matrix of
P
21:
ω
=
t
*/
22:
m
←{
m
,
1
/ω
}
(
m
,I)
←
reverse-sort
(
m
)
/* sort
m
in descending order, and put the corresponding indices in
I
*/
23:
24:
I
←
I(
1:
n)
/* Use the most similar previous minority class examples to augment
S
(t)
to achieve
desired class ratio
f
.*/
←
L(
S
t
+
Q
t
(I ))
25:
h
t
26:
Q
←{
Q
,
P
t
}
27:
if
d
=
SERA
then
return
h
t
final
=
h
t
for predicting class label of instance
x
within
(t)
28:
T
29:
else
30:
H
←{
H
,h
t
}
31:
W
←{}
32:
for
i
:1
→
t
do
(
x
j
,y
j
)
∈
S
t
(
1
f
y
j
i
1
|
S
t
|
(
x
j
))
2
33:
e
i
=
−
34:
W
←{
W
,
log 1
/e
i
}
return
Composite hypothesis
h
(t)
final
for predicting class label of any instance
x
j
35:
in testing
data set
T
t
is
H
×
W
, i.e.,
h
(t)
t
final
(
x
j
)
f
i
(
x
j
)
=
argmax
c
∈
Y
w
i
×
(7.19)
i
=
1
Search WWH ::
Custom Search