Criação de um método para reduzir a dimensão da hierarquia de classes durante o treinamento do algoritmo MHC-CNN

Abstract

Data classification is a common task in Data Mining. Some classification problems need to take into account the data hierarchical taxonomy, establishing an order between classes, those are called hierarchical classification problems. This work presents a method to reduce the dimension of the hierarchical classes in the MHC CNN hierarchical classifier training phase. The method is called HNR and seek to improve the algorithm MHC-CNN. The proposed method is called HNR and attempt to improve the MHC-CNN. To evaluate the method, ten protein functions datasets is used and the results compared to MHC-CNN’s results. Preliminary studies show that the HNR is an alternative that can be used for global hierarchical classification.