Algorithm 1. Training Decision Tree

Input: dataset $\mathit{D}$

Output: the decision tree

1. Initialize an empty tree

2. Generate processed training dataset $D_{train}$

3. repeat

4. For each attribute $a\in D$

5. Compute the gain_ratio of $a$

6. End

7. choose the best split attribute $a_{sp}$ based above computed criteria

8. create a decision node and attach this node to the corresponding branch of the tree T

9. partition the dataset to subdatasets based on $a_{sp}$

10. for each subdatasets $D_i$

11. Repeat same operation from 3 to 12.

12. End

13. until $D_i$ is pure or size of $D_i$ less than minimum or the algorithm reaches enough iterations

14. return $T$.