Recurrence microstates for machine learning classification.

Autor:	Spezzatto GS; Department of Physics, Federal University of Paraná, 81531-980 Curitiba, Brazil., Flauzino JVV; Department of Physics, Federal University of Paraná, 81531-980 Curitiba, Brazil., Corso G; Biophysics and Pharmacology Department, Federal University of Rio Grande do Norte, 59078-900 Natal, Rio Grande do Norte, Brazil., Boaretto BRR; Institute of Science and Technology, Federal University of São Paulo, 12231-280 São José dos Campos, São Paulo, Brazil., Macau EEN; Institute of Science and Technology, Federal University of São Paulo, 12231-280 São José dos Campos, São Paulo, Brazil., Prado TL; Department of Physics, Federal University of Paraná, 81531-980 Curitiba, Brazil.; Department of Physics, University Rey Juan Carlos, Móstoles, 28933 Madrid, Spain.; Interdisciplinary Center for Science, Technology and Innovation CICTI, Federal University of Paraná, 81531-980 Curitiba, Brazil., Lopes SR; Department of Physics, Federal University of Paraná, 81531-980 Curitiba, Brazil.; Interdisciplinary Center for Science, Technology and Innovation CICTI, Federal University of Paraná, 81531-980 Curitiba, Brazil.; Potsdam Institute for Climate Impact Research-Telegraphenberg, 14473 Potsdam, Germany.
Jazyk:	angličtina
Zdroj:	Chaos (Woodbury, N.Y.) [Chaos] 2024 Jul 01; Vol. 34 (7).
DOI:	10.1063/5.0203801
Abstrakt:	Recurrence microstates are obtained from the cross recurrence of two sequences of values embedded in a time series, being the generalization of the concept of recurrence of a given state in phase space. The probability of occurrence of each microstate constitutes a recurrence quantifier. The set of probabilities of all microstates are capable of detecting even small changes in the data pattern. This creates an ideal tool for generating features in machine learning algorithms. Thanks to the sensitivity of the set of probabilities of occurrence of microstates, it can be used to feed a deep neural network, namely, a microstate multi-layer perceptron (MMLP) to classify parameters of chaotic systems. Additionally, we show that with more microstates, the accuracy of the MMLP increases, showing that the increasing size and number of microstates insert new and independent information into the analysis. We also explore potential applications of the proposed method when adapted to different contexts. (© 2024 Author(s). Published under an exclusive license by AIP Publishing.)
Databáze:	MEDLINE
Externí odkaz:	Zobrazit plný text záznamu