Feature selection is a widespread preprocessing step in the data mining field. One of its purposes is to reduce the number of original dataset features to improve a predictive model's performance. Despite the benefits of feature selection for the classification task, as far as we are aware, few studies in the literature address feature selection for hierarchical classification context. This paper proposes a novel feature selection method based on the General Variable Neighborhood Search metaheuristic, combining a filter and a wrapper step, wherein a global model hierarchical classifier evaluates feature subsets. We used twelve datasets from protein and image domains to perform computational experiments to validate the proposed algorithm effect on classification performance when using two global hierarchical classifiers proposed in the literature. Statistical tests showed that using our method as a feature selection led to a predictive performance that is consistently better or equivalent to that obtained by using all features, with the benefit of reducing the number of features needed, which justifies its efficiency for the hierarchical classification scenario.

Hoje vivemos uma mudança de paradigma no setor financeiro, com forte redução das agências bancárias físicas e aumento de serviços online. Contudo, a facilidade de abertura de contas digitais propiciada por esta mudança de paradigma também tem levado a um aumento nos casos de fraude. Este trabalho apresenta o problema de detecção de fraude financeira sob uma nova taxonomia e, também, investiga técnicas de classificação hierárquica para a tarefa. A abordagem hierárquica global (CLUS-HMC), em que toda a hierarquia de classes é considerada pelo classificador, resultou em melhores valores de Recall para as classes fraudulentas (33.31% para classe E e 35.09% para classe F), indicando um caminho de pesquisa promissor.