Method of hybrid logical classification trees based on group selection of discrete features

The paper considers the problems of automating the construction of classification trees based on the scheme of branched feature selection. The object of research is classification trees. The subject of research is methods, algorithms, and schemes for constructing classification trees. The aim of this work is to build an effective method (scheme) for synthesizing classification tree models based on a group assessment of the importance of discrete features within a branched attribute selection. A method for constructing classification trees is proposed, which for a given training sample determines the individual information content (importance) of groups of features (and their combinations) in relation to the initial value of the classification function (data from the training sample). The developed logical tree method, when constructing the next node of the classification tree, tries to identify a group of the most closely interrelated discrete features, this reduces the overall structural complexity of the model (the number of levels of the classification tree), speeds up calculations when recognizing objects based on the model, and also increases the generalizing properties of the model and its enterprise. The proposed scheme for selecting groups of discrete traits allows using the constructed decision tree to assess the informative value (importance) of traits. The developed classification tree method is implemented programmatically and studied when solving the problem of classifying discrete objects represented by a set of features. The conducted experiments confirmed the operability of the proposed mathematical support and allow us to recommend it for use in practice in solving applied problems of classification of discrete objects based on logical classification trees. Prospects for further research may consist in creating a modified method of the logical classification tree by effectively iterating and evaluating sets of elementary features based on the proposed method, optimizing its software implementations, and experimentally studying the proposed method on a wider set of applied problems.