A Novel Hybrid Feature Selection Algorithm for Hierarchical Classification

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