An improved ordinal decision tree induction algorithm

For the existing ordinal decision tree induction algorithms, ranking mutual information between the conditional attributes and the decision attribute is employed as a heuristic to select the candidate attribute, while the correlation among the conditional attributes is not considered. To solve this problem, this paper proposes an improved ordinal decision tree induction algorithm. The candidate attributes selected with the proposed algorithm not only maximize the ranking mutual information between the candidate attributes and the decision attribute, but also minimize the ranking mutual information between the candidate attributes and the selected conditional attributes on the same branch. Taking into account the redundancy of condition attributes that can avoid repeating selection of the same condition attributes, we perform experiments which show that the ideas of the proposed method can really reflect the nature of the ranking mutual information. Compared with the existing algorithms, the proposed algorithm can improve the testing accuracy.