Hierarchical Single Label Classification: An Alternative Approach

In this paper an approach to multi-class (as opposed to multi-label) classification is proposed. The idea is that a more effective classification can be produced if a coarse-grain classification (directed at groups of classes) is first conducted followed by increasingly more fine-grain classifications. A framework is proposed whereby this scheme can be realised in the form of a classification hierarchy. The main challenge is how best to create class groupings with respect to the labels nearer the root of the hierarchy. Three different techniques, based on the concepts of clustering and splitting, are proposed. Experimental results show that the proposed mechanism can improve classification performance in terms of average accuracy and average AUC in the context of some data sets.