The Use of the Label Hierarchy in Hierarchical Multi-label Classification Improves Performance

We address the task of learning models for predicting structured outputs. We consider both global and local approaches to the prediction of structured outputs, the former based on a single model that predicts the entire output structure and the latter based on a collection of models, each predicting a component of the output structure. More specifically, we compare local and global approaches in terms of predictive performance, learning time and model complexity. Moreover, we discuss the interpretability of the obtained models. We evaluate the predictive performance of the considered approaches on six case studies from three domains: ecological modelling, text classification and image classification. Finally, we identify the properties of the tasks at hand that lead to the differences in performance.