Information Technology Reference
In-Depth Information
Algorithm 7.11
Decision tree generating algorithm GSD
Input: Training set
Output: Decision tree, rule set
1. Select standard CR from Gain, Gain Ratio and ASF;
2. If all data items belong to one class, then decision tree is a leaf labeled with the class
sign;
3. Else, use optimum test attribute to divide data items into subset;
4. Recursively call step 2
、
3 to generate a decision tree for each subset;
5. Generate decision tree and transform to rule set.
7.8.9
Experiment results
BSDT algorithm uses following dataset (Table 7.3).
Table 7.3 Experiment dataset
#example
number of test
set
#training example
number
Dataset
#attribute number
#class number
anneal
breast-cancer
credit
genetics
glass
heart
hypo
letter
sonar
soybean
voting
diabetes
898
699
490
3,190
214
1,395
2,514
15,000
208
683
300
768
38
10
15
60
9
16
29
16
60
35
16
8
5
2
2
3
6
2
5
26
2
19
2
2
-
-
200
-
-
-
1,258
5,000
-
-
135
-
Parts of the dataset come from UCI test database (ftp://ics.uci.edu/pub/
machine-learning-database). Following measure is used to test dataset.
Using random selecting approach to divide dataset DS without given dataset
(such as: anneal, breast-cancer, etc) into test set TS and learning set LS, and let
TS = DS×10%;
