DOI: https://doi.org/10.15407/pp2018.02.171

Characteristics for choice of models in the ansables classification

O.V. Barmak, Yu.V. Krak, E.A. Manziuk

Abstract


The analysis of researches and approaches of practical application of ensembles is carried out and the characteristic factors of influence on a combination of models are determined. Factors are determinative and inherent in combinations of models. The necessity of using patterns of models that are characteristic only for ensembles is substantiated. The characteristic features of ensembles are established and the necessity of developing specific characteristics of models in terms of a combination of solutions by definition of their characteristic features is determined. These attributes, namely precision and distinction have a decisive influence on the choice and applicability of solutions in ensembles and allow to choose the most effective combination. We propose to use such a characteristic of the models in the ensemble as distinction of a certain rank by precision parameter. These characteristics of the models allow them to be chosen and characterize the model in the ensemble. It applies only in the case of a combination of models and indicates the distinction between one model and another and takes into account the precision of the model. This approach allows you to define models of different nature, to determine the depth of model distinction and allows it to be combined with other well-known characteristics of classifiers. Its peculiarity consists in the fact that it allows to evaluate the use of solutions in the ensemble and to carry out the selection of models.

Problems in programming 2018; 2-3: 171-179


Keywords


ensembles; ensemble size; distinction; precision

Full Text:

PDF (Ukrainian)

References


Brownlee J. (2014) A Data-Driven Approach to Choosing Machine Learning Algorithms. [Online] September 29th 2014. Available from: Machinelearningmastery.com https://machinelearningmastery.com/a-data-driven-approach-to-machine-learning/. [Accessed: 30 January 2018].

Ren S., Liao B., Zhu W., Li K. (2018) Knowledge-maximized ensemble algorithm for different types of concept drift. Information Sciences. 430-431 (3). P. 261-281.

https://doi.org/10.1016/j.ins.2017.11.046

Hsu C-H., Shi X., Salapura V. (2014) 11th IFIP International Conference on Network and Parallel Computing (NPC), 18-20 September 2014. Ilan, Taiwan. Springer. LNCS-8707. P. 333-346, 2014. Network and Parallel Computing.

Diez-Pastor J.F., Rodríguez J.J., García-Osorio C. & Kuncheva L.I. (2015) Diversity techniques improve the performance of the best imbalance learning ensembles. Information Sciences. 325. P. 98-117.

https://doi.org/10.1016/j.ins.2015.07.025

Löfström T. (2015) On Effectively Creating Ensembles of Classifiers: Studies on Creation Strategies, Diversity and Predicting with Confidence. A Thesis Submitted in partial fulfilment of the Requirements of Stockholm University for the Degree of Doctor of Philosophy. Stoke-on-Trent: Stockholm University.

Hamed R.B., Fazli C. (2017). Less Is More: A Comprehensive Framework for the Number of Components of Ensemble Classifiers. IEEE Transactions on Neural Networks and Learning Systems. [Online] 14(8), September 2017 USA: IEEE. P. 1-7. Available from: https://arxiv.org/pdf/1709.02925.pdf [Accessed 30/01/2018].




DOI: https://doi.org/10.15407/pp2018.02.171

Refbacks

  • There are currently no refbacks.