Information Technology Reference
In-Depth Information
IB3 is the improvement of IB2 that employs a “selective utilization filter”
to determine which of the saved instances should be used to make
classification decisions in such a way that IB3 is insensitive to noise.
Algorithm 5.3: IB3
1.
CD
∅
//CD=Concept description
2. for each
x
∈ Training Set do
3. for each
y
∈CD do
4. sim[
y
]←similarity(
x,y
)
5.
if ∃{
y
∈CD|acceptable (
y
)} then
6.
y
max
←some acceptable
y
∈CD with maximal sim[
y
]
7.
else
8.
i
←a randomly-selected value in [1,|CD|]
9.
y
max
←some y∈CD that is the
i
-th most similar instance to x
10.
if class(
x)
class(y
max
)
11.
then
12.
classification←correct
13.
else
14.
classification←incorrect
15.
CD←CD
∪
{
x
}
16.
for each
y
∈CD do
17.
if sim [
y
]
sim[
y
max
]
18.
then
19.
update
y
's classification record
20.
if
y
's record is significantly poor
21.
then CD←C-{
y
}.
The IB3 algorithm maintains a classification record, e.g., the number of
correct and in correct classification attempts, with each saved instance. A
classification record is a summary of an instance's classification performance
on the current training instances and hints the performance in the future.
Meanwhile, IB3 employs a significance test to determine which instances are
good classifiers and which ones are noisy. The noisy instances are discarded
from the concept description. For each training instance i, all classification
records are updated for instances that are at least as similar as i's most similar
acceptable neighbor.
