4.1. Attribute Reduction of Incomplete Decision Tables
To verify the effectiveness of the proposed algorithm, an incomplete decision table (as shown in
Table 4) in [
14] is adopted. Next, we apply our heuristic algorithm to calculate the attribute reduction of
Table 4.
Example 2 ([
14]).
Given an incomplete decision table (as shown in Table 4), , . Step 1. Firstly, we apply a Boolean matrix to calculate
. According to
Table 4, the Boolean relation matrices of
, and
can be acquired as below.
According to matrix , it is obvious that , , , , , , , , , , , .
According to matrix , we have , , , , , , , , , , , , then .
Secondly, it is easy to obtain , , , , , , , , , , , .
Finally, according to Definition 7, we can obtain a binary similarity matrix (as shown in
Table 5).
Step 2. According to Theorem 1, we can obtain .
Step 3. Test whether
is a reduction of
C. Similar to the method of calculating
,
. Thus, we delete the columns where attributes
,
and
are located in
Table 5 to obtain
Table 6. Initialization:
.
Step 4. According to
Table 6, we have
. Calculate the
of attributes in
Table 6:
,
,
,
,
.
Step 5. Obviously,
of
is the smallest, so delete the column where
is located and the row with value “1” in this column, and obtain
Table 7, then
. Because
Table 7 has “1”, go to Step 4. Calculate the
of attributes in
Table 7:
,
,
,
. It is obvious that
of
and
are the smallest. If we delete the column where
is located and the row with value “1” in that column to obtain
Table 8, then
.
Table 8 has “1”, then go to Step 4 again. Perform Step 4 again. Calculate the
of attributes in
Table 8:
,
,
. Then, delete the column where
is located and the row with value “1” in that column, and we have
.
does not have “1”, so then go to Step 6.
Step 6. .
In
Table 7, select the column where
is located and the row with value “1” in that column to obtain
Table 9. Then, according to
Table 9, delete the column where
is located and the row with value “1” in that column, and we can obtain
, then delete
and in the last row, we also obtain
.
In order to verify the effectiveness of our algorithm, it is necessary to test whether T is a reduction of C. When , . When , we have , and there are no such that . Thus, is a reduction of attributes set C.
Moreover, the reduction result in reference [
14] is
. It is obvious that our result contains five attributes, and their result contains six attributes. What is more, it can be proved that our result is also a reduction of
C from the point of literature [
14]. The purpose of attribute reduction is to classify objects with as few attributes as possible. From the discussion above, our algorithm is not only effective but also efficient.