Detecção de periodicidade em dados multi frequência de núcleos ativos de galáxias com aprendizagem de máquina (XGBoost)

• Ano de defesa: 2019
• Autor(a) principal: Soltau, Samuel Bueno
• Orientador(a): Botti, Luiz Claudio Lima
• Banca de defesa: Pamboukian, Sérgio Vicente Denser, Wuensche, Carlos Alexandre, Boas, José Williams dos Santos Vilas
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Presbiteriana Mackenzie
• Programa de Pós-Graduação: Engenharia de Materiais e Nanotecnologia
• Departamento: Escola de Engenharia Mackenzie (EE)
• País: Brasil
• Palavras-chave em Português: núcleos ativos de galáxias; periodicidade; aprendizado de máquina; XGBoost
• Área do conhecimento CNPq: CNPQ::CIENCIAS EXATAS E DA TERRA::GEOCIENCIAS
• Link de acesso: http://dspace.mackenzie.br/handle/10899/25803

One ofthe most interesting aspects of compact radio-sourcesit’svariability.This study use a Machine Learning method called eXtreme Gradient Boosting(XGBoost) to investigate PKS 1921-293 and BLLac 2200+420 periodicities,through flux density treatment and analysis ofthe light curves,data sets from the University of Michigan Radio Astronomy Observatory(UMRAO),acquiredat 4.8 GHz, 8.0 GHz, and 14.5 GHz. The time period cover edto PKS1921-293,respective lyin these frequencies were from1980 to 2011,from 1975 to 2011 and from 1976 to 2011 and to BLLac 2200+420,respectively in the same frequencies were 1977 to 2012,from 1968 to 2012 and from 1974 to 2012. was used int wost eps ofthe method:time series regularization in order to obtaina regularly-spaced temporal flux density data,and to classification off luctuations in the flux density of thet wo objects in order to detect the occurrence of out burst periodicity.For each methodological stage,feature trainings were carried out with different purposes.In time series regularization,XGBoost wasable to produce regularly time serie for light curve with daily data.In periodi city search,XGBoost identified data segment son which the occurrence of out burst was estimated.Forboth BLLac2200+420andPKS1921-293,the periodi cities found were compared to those in the literature.XGBoost library robust ness provide de Xtreme Gradient Boosting the opportunity to use a Machine Learning based methodology on radio data and extract information with strategies quite different from those traditionally used to treat irregularly time series and toobta in periodicity through the classification o frecurrent events