Hierarchical classification is a variant of multidimensional classification where the classes are arranged in a hierarchy and the objective is to predict a class, or set of classes, according to a taxonomy. Different alternatives have been proposed for hierarchical classification, including local and global approaches. Local approaches are prone to suffer the inconsistency problem, while the global approaches tend to produce more complex models. In this paper, we propose a hybrid global-local approach inspired on multidimensional classification. It starts by building a local multi-class classifier per each parent node in the hierarchy. In the classification phase all the local classifiers are applied simultaneously to each instance resulting in a most probable class for each classifier. A set of consistent classes are obtained, according to the hierarchy, based on three novel alternatives. The proposed method was tested on three different hierarchical classification data sets and was compared against state-of-the-art methods, resulting in significantly superior performance to the traditional top-down techniques; with competitive results against more complex top-down classifier selection methods.
展开▼
