Hierarchical Classification Using a New Hybrid Feature Selection Algorithm

Abstract

A common preprocessing step in the data mining industry is feature selection. Reducing the quantity of original dataset characteristics is one of its goals in order to enhance the accuracy of a prediction model. To the best of our knowledge, very few research in the literature address feature selection for the context of hierarchical classification, despite the advantages of feature selection for the classification problem. The general variable neighbourhood search metaheuristic is used to support the innovative feature selection approach that is proposed in this research. The method combines a filter step and a wrapper phase, and a global model hierarchical classifier is used to assess feature subsets. We conducted computational tests to verify the impact of the suggested approach on classification performance while employing two proposed global hierarchical classifiers, using various datasets from the proteins and pictures domains. in the written word. According to statistical testing, our feature selection strategy consistently produced prediction results that were superior to or on par with those achieved by employing all features while using fewer features, which supports its efficacy in the context of hierarchical categorization.