Tree-based dynamic classifier chains

Abstract Classifier chains are an effective technique for modeling label dependencies in multi-label classification. However, the method requires a fixed, static order of labels. While theory, any is sufficient, practice, this has substantial impact on quality final prediction. Dynamic classifier denote idea that each instance to classify, which labels predicted dynamically chosen. The complexity naïve implementation such approach prohibitive, because it would require train sequence classifiers every possible permutation To tackle problem efficiently, we propose new based random decision trees can select ordering We show empirically dynamic selection next improves over use under otherwise unchanged tree model. In addition, also demonstrate alternative extreme gradient boosted trees, allows more target-oriented training chains. Our results variant outperforms and other tree-based classification methods. More importantly, strategy considerably speed up