HCAIM: a discretizer for the hierarchical classification scenario applied to bioinformatics datasets

Abstract

Discretization is one of the stages of data preprocessing that has been the subject of research in several works related to flat classification. Despite the importance of data discretization for a classification task, to the best of our knowledge, when it comes to the hierarchical classification scenario, where the classes to be predicted are organized according to a hierarchy, there are no discretization methods in the literature that take class hierarchy into account. The development of discretization methods capable of dealing with class hierarchy is extremely important to enable the use of global hierarchical classifiers that require discrete data. Therefore, in this work, we fill this gap by proposing and evaluating a supervised discretization method for the hierarchical classification context. Experiments with 17 bioinformatics datasets using a global hierarchical classifier showed that the proposed method allowed the classifier to achieve predictive performance superior to those obtained when other unsupervised discretization methods were used.