Research of Methods of Identifying the Computer Systems State Based on Bagging Classifiers

EasyChair Preprint 9397

Abstract

Peculiarities of constructing ensemble bagging classifiers for identifying the state of a computer system under conditions of noisy data are studied. Decision trees and multilayer perceptron were used as basic classifiers. It was found that the accuracy of the bagging algorithm with decision trees as basic classifiers with standard settings ranges from 84.4% to 88.7%. The use of Bootstrap algorithms for the formation of data samples: Pasting, Bootstrapping, Random Subspace, Random Patches Ensemble and the selection of the number of basic classifiers in the ensemble made it possible to increase the classification accuracy to 90.2%. The following parameters were added to improve the accuracy of bagging classifiers based on the multilayer perceptron: the algorithm for forming data samples, the number of basic classifiers in the ensemble, the function of optimizing the neural network, the function of activating hidden layer, size of hidden layers. The recommendation was made to choose the value of the analyzed parameters for the creation of bagging ensembles with multilayer perceptrons, which made it possible to increase the accuracy of computer system identification up to 92.2%. The obtained results have further practical significance and can be used in improving the methods of identifying the state of computer systems.

Keyphrases: Bagging, Multilayer Perceptron, computer system, decision trees, machine learning, state identification

@booklet{EasyChair:9397,
  author    = {Svitlana Gavrylenko and Oleksii Hornostal and Viktor Chelak},
  title     = {Research of Methods of Identifying the Computer Systems State Based on Bagging Classifiers},
  howpublished = {EasyChair Preprint 9397},
  year      = {EasyChair, 2022}}