Overview and analysis approaches for classifying with a heterogeneous set of information features

Abstract

Background. Data classification is an important part of data processing. In the modern world, objects that need to be classified are often heterogeneous - they have information features of different types: numeric, textual, graphical, graph, multimedia. This study is devoted to the review and analysis of a number of existing methods for classification of objects with heterogeneous space of information features. Original approach based on the generation of raster grapho-chromatic maps was proposed. Materials and methods. In this study, the problem of neural network classification of objects with a heterogeneous space of information features is formulated, taking into account the possibility of controlling their quantitative and qualitative parameters without the need to retrain the neural network. Modern classification methods were considered and their features were analyzed. Results and conclusions. The main cons of existing methods for classifying heterogeneous objects were identified and a new approach was proposed, based on the generation of a universal graphic code, with the help of which heterogeneous features will be reduced to a single graphic representation for further neural network classification.

Keywords

• classification
• methods of classification
• heterogeneous data
• machine learning
• neural networks
• graph-chromatic map