CONSTRUCTION OF MULTILEVEL ENSEMBLES IN MACHINE LEARNING PROBLEMS

Abstract

The approach to solving the problems of machine learning and motivation on the basis of rich heterogeneous ensembles is examined. The approach allows you to change forecast estimates based on one-hour changes in variance and variance based on a variety of diverse heterogeneous ensembles of forecast models. The main features and changes in the creation of ensembles are examined. An analysis was carried out for the purpose of the machine learning algorithm. It consists of three components: noise, displacement and dispersion. The warehouse components have been tracked and identified. The process of creating a heterogeneous ensemble is examined. The most extensive methods of aggregation of forecast values ​​have been analyzed: bagging, boosting and staking. The choice of type of models for the creation of a heterogeneous rich ensemble structure is given. The courtyard architecture of the classification system based on the methods of Staking and Bagging has been proposed. To create a rich ensemble of models based on various basic methods, an algorithm has been developed. The results of the robotic ensembles were analyzed. The effectiveness of a large number of heterogeneous ensembles with the highest specification of classification has been demonstrated.