Previsão da quantidade de classes em um problema de classificação hierárquica multirrótulo

Abstract

In hierarchical multi-label classification problems, each instance may be associated with one or more labels simultaneously belonging to a hierarchical level subclass or superclass. This classification is performed through learning techniques that from a trained database defines the set of labels that an instance is associated. However, an existing multi-label problem in rating refers to classifiers used in existing techniques which do not provide the number of labels to be defined for new instances presented to the classification model. In the current literature of Learning Machine and Data Mining, these classifiers only define in the training phase the set of labels of entry example with amount already preset in the training phase. Therefore, in this study will be investigated and adapted classification techniques able to predict the number of labels to be associated with a new instance, such that it is contained in a hierarchical and multi-label entries.