Artificial intelligence

Scientific journal

ISSN 2710-1673

ONLINE: ISSN 2710-1681

Search by:

Year of publication
Author name
Paper title

Principles and approaches to the formation of ansambles based on analysis results

Barmak2 O.1, Krak U.2, Manziuk E.1, Kulias A.4
1 Khmelnytskyi National University
2 Taras Shevchenko National University of Kyiv
4 Glushkov Institute of Cybernetic of NAS of Ukraine

Full text (PDF)

UDC: 004.912
Publication Language: Ukrainian
Stuc. intelekt. 2018; 23(3):62-69

Abstract: The article considers principles of the formation and application of classifiers' ensembles. The advantages of using the approaches of aggregation of classifier solutions are considered. An analysis of the principles of data redistribution with the extended ensemble made it possible to formulate requirements for the classifiers of the ensemble. On the basis of what was found the conditions for the selection of classifiers and the criteria for their adequacy. On the basis of the sufficiency criteria, the applicability conditions for the principle of aggregation for boundary sets were established. An analysis of the ambiguity of decision making for symmetric sets of classifiers is carried out. The conditions for forming a set of classifiers and the criteria for its adequacy for solving the recognition problem are determined. It is established that further expansion of the set of classifiers over the sufficiency criterion brings classification errors to the set of correct solutions of all classifiers. The extension of the classifier set allows us to form a set of connections that is a Pascal triangle and to analyze the marginal redistribution of data in the process of increasing the ensemble.

Keywords:

References:

  1. Mueller, D. (2003). Public Choice III.Cambridge: University of Cambridge Press.
  2. Gerecke, U., Sharkey, N. E., Sharkey, A. J.(2003). Common evidence vectors for selforganized ensemble localization,Neurocomputing, 55(3), P. 499-519.
  3. Schapire, R.E., (1990). The strength of weaklearnability, Machine learning, 5(2), P. 197-227.
  4. Zhu, M (2015). Use of majority votes in statisticallearning. Wiley Interdisciplinary Reviews:Computational Statistics, 7(6), P. 357-371.
  5. Vardeman, S.B., Morris, M.D. (2013), “MajorityVoting by Independent Classifiers can Increase ErrorRates,” The American Statistician, 67, P. 94-96
  6. Vardeman, S.B., Morris, M.D. (2013), “MajorityVoting by Independent Classifiers can IncreaseError Rates,” The American Statistician, 67,P. 94–96: Comment by Stuart Baker, Jian-LunXu, Ping Hu and Peng Huang and Reply TheAmerican Statistician, 2014, 68(2), P. 125-126
  7. Manziuk, E.A., Barmak, A.V., Krak, Y.V.,Kasianiuk, V.S. (2018). Opredelenieinformatsionnogo yadra pri klassifikatsiidokumentov. Problemyi upravleniya iinformatiki, 50(4), P. 25-34.DOI: 10.1615/JAutomatInfScien.v50.i4.30

View full text (PDF)