Information Technology Reference
In-Depth Information
frequency propagating. At the same time, it realizes generalization consistency
examination, data noise eliminating, automatic concept level generating, etc.
Algorithm 7.6
Classification guided attribute oriented inductive algorithm
CGAOI
InputPrimitive training instance subset
E
0
, current attribute
A
; concept
level tree
T
, appointed concept level tree
L
; appointed attribute controlling threshold
Y
;
current concept level, current attribute controlling threshold
Y
0
, attribute set
A
0
.
Output: appointed concept level training example subset
E
, attribute set A, frequency
C
T
.
1. Call algorithm GCCC to generalize consistency check and noise elimination, return
Ret1;
2. If generalization consistency check fails, then return failure;
3. Do concept improving operation under the control of
Y
and
L
;
4. Attribute elimination and frequency propagation;
5. Return success.
Theorem 7.2
The representative bias shift will be stronger and more accurate
through algorithm CGAOI processing.
2. Preprocessing algorithm PPD
In the algorithm, we take specific attribute value of database as bound of
generalization and specification. When attribute controlling threshold larger than
current attribute value (or level lower than current level), it calls generalization
procedure; on the contrary, it calls specialization procedure to reduce primitive
training instances subset to corresponding concept level.
Algorithm 7.7
Preprocessing algorithm PPD
0
; concept level forest
F
;
concept level database
D
; appointed concept level
L
; appoint attribute controlling
threshold
Y
; current concept level; current attribute controlling threshold
Y
0
; attribute set
A
Input: Primitive training instance subset
E
0
;
Output: Appointed concept level training example subset
E
; attribute set
A
;
0
;
1. Do operation to each attribute
A
i
in attribute set
A
2. Whether concept level tree
F
i
of
A
i
is empty;
3. if it is empty, then call algorithm AGCH to generate concept level tree
F
i
of
A
i
automatically, return Ret1;
4. Ret1= -1, go to step 1; // Ai have no concept level tree
5. If (
Y
=
Y
0
∩
L
=
L
0
), then go to step 1;
0
), then call generalization algorithm CGAOI, return Ret2;
7. Ret2 = -1go to step 1; // if failthen abandon generalization
8. Ret2 = 0go to step 11; // generalization success
0
∪
L
>
L
6. If (
Y
<
Y
