Weighted Fuzzy Decision Tree for Multi-Label Classification

Multi-label classification is a pattern recognition task to assign one or more class labels to each observed instance. In multi-label classifiers, there are complex decision boundaries and using simple and powerful classifiers such as decision trees can satisfy most of the requirements due to its capability of space partitioning from general to specific points of view. In addition, merging it with fuzzy theory enhances the interpretability of the algorithm and smoothens the boundaries. In this paper, a novel fuzzy decision tree is proposed and customized for multi-label classification. In fuzzy classifiers, more than one leaf may deal in reasoning. A weighting mechanism is also proposed to adjust the importance of the leaves in voting for classification. The most important characteristic of the proposed system is that it considers the label dependencies in both stages of tree construction and weighting in order to improve the performance of multi-label classification. An empirical analysis, conducted on 5 multi-label datasets, shows that the proposed fuzzy decision tree performs better than other tree-based approaches in the literature. Further experimental results show that, not only the proposed weighting method significantly outperforms other algorithms, but also it boosted the classification ability of the initial fuzzy decision tree.