HCGA: A genetic algorithm for hierarchical classification

Hierarchical classification (HC) is a specialization of the well-known flat classification task. The main difference among them is that in HC examples have to be assigned to classes organized in a previously defined class hierarchy, while in traditional flat classification no class order is imposed. There are two main approaches commonly used to tackle HC: the top down or local approach, which is classifier independent, and the big-bang or global approach, which usually is the product of a modification of a well-known flat classifier. Although evolutionary algorithms have been successfully applied to flat classification, they are underexplored in HC. In this direction, this paper pro poses HCGA (Hierarchical Classification Genetic Algorithm), a method that takes both local and global information into account. HCGA uses a top-down approach for building a classification model and also for classifying new examples. This is in contrast with current top-down methods, which make use of this strategy only for test, using flat classifiers for training models. The method was applied to four GPCR (G protein-coupled receptor) activity datasets, obtaining results statistically equal or better than five baseline classifiers run using a top-down approach.