A Procedure for Classifying Objects with a Semantic Hierarchy of Features

Abstract

A procedure for classifying objects with a hierarchical structure of relations (semantics) of features that takes into account their modalities is proposed. The concepts of semantics of features and their modalities are explained prior to the description of this procedure. A three-level model with a semantic hierarchy of features (objects—meta-features—subfeatures of objects) is considered. Meta-features are interpreted as semantic generalizations of the related subfeatures of objects. An important stage of the proposed procedure is the aggregation of the lower level subfeatures, taking into account their semantic connection with the meta-features. Aggregation leads to a significant reduction in the dimension of the original classification problem, which is now solved in terms of values of the aggregation function. As an example, the Dermatology sample from the well-known UCI Machine Learning repository is considered. This example shows that despite a considerable imbalance of the Dermatology sample, the results yielded by the proposed procedure are quite comparable with the best results of some well-known algorithms obtained on this sample.