The decision tree is one of the most popular data mining tools for solving classification and prediction problems. Due to its computational speed and interpretability, the decision tree for a single objective has been successfully used in various fields. In practice, however, data sets with multiple objectives are frequently encountered, and therefore, the traditional decision tree needs to be extended to handle such multi-objective data.
This thesis proposes the misclassification ratio as a new split criterion for the multi-objective decision tree. The proposed split criterion is then compared with other existing split criteria using real data. Experimental results indicate that the performance of the misclassification ratio is better or at least not worse than other existing split criteria in terms of prediction accuracy, interpretability, computational speed, and scalability. In addition, the proposed misclassification ratio is conceptually simple, and is believed to be a promising alternative to the existing split criteria for multi-objective decision trees.