METHODS OF SYNTHESIS OF MODELS OF QUANTITATIVE DEPENDENCIES IN THE BASIS OF TREES OF REGRESSION, REALIZING CLUSTER-REGGRESSION APPROXIMATION BY PRECEDENTS

Abstract

Context. To make decisions in technical applications, it is usually necessary to have a model that allows you to predict the state of a managed object or process. The object of the study is the process of building dependency models by use cases. The subject of the study are the methods for constructing quantitative dependencies based on cluster-regression approximation precedents. Objective. The aim of the paper is to simplify cluster regression approximation models by indirectly implementing cluster analysis in the process of model building. Method. A tree-based cluster-regression approximation method is proposed which, for a given training sample, constructs a tree for hierarchical clustering of instances whose leaf nodes correspond to clusters, for each cluster, constructs a particular model of dependence on instances of the training sample that fall into the cluster, in order to provide the least complexity of the model and uses the set the most informative features of the shortest length. This allows to ensure an acceptable accuracy of the model, high levels of interpretation and generalization of data, to reduce the complexity of the model, and to simplify its implementation in the sequential organization of calculations. Results. The software that implements the proposed method of tree-like cluster-regression approximation is developed. The developed method and the software implementing it are investigated in solving practical problems of prediction. The conducted experiments confirmed the working capacity of the developed software and allow to recommend it for use in practice. Conclusions. Unlike traditional methods of regression model constructing that build a model based on a function form that is uniform for the entire feature space, the proposed method forms a hierarchical combination of particular models. Unlike the wellknown methods of regression tree constructing whose leaf nodes contain averaged values of the output feature for clusters, the proposed method forms a tree consisting of particular models for clusters, which allows to ensure greater accuracy of the model.