Abstract:

An improved ordinal decision tree algorithm was proposed. The extended attributes selected with the proposed algorithm maximized the ranking mutual information between the candidate attributes and the decision attribute, and also minimized the ranking mutual information between the candidate attributes and the selected conditional attributes on the same branch. The experimental results showed that  the correlation to be taken account among the conditional attributes could  avoid to  selecte  the same one, and the ideas of the proposed method could really reflect the nature of the ranking mutual information. The proposed algorithm could improve the test accuracy compared with the existing algorithms.

Key words: ranking mutual information, ordinal classification, ranking entropy, correlation of attribute, decision tree