Uso e estabilidade de seletores de variáveis baseados nos pesos de conexão de redes neurais artificiais
Autor(a) principal: | |
---|---|
Data de Publicação: | 2021 |
Tipo de documento: | Tese |
Idioma: | por |
Título da fonte: | Repositório Institucional da UFG |
Texto Completo: | http://repositorio.bc.ufg.br/tede/handle/tede/11396 |
Resumo: | Artificial Neural Networks (ANN) are machine learning models used to solve problems in several research fields. Although, ANNs are often considered “black boxes”, which means that these models cannot be interpreted, as they do not provide explanatory information. Connection Weight Based Feature Selectors (WBFS) have been proposed to extract knowledge from ANNs. Most of studies that have been using these algorithms are based on just one ANN model. However, there are variations in the ANN connection weight values due to the initialization and training, and consequently, leading to variations in the importance ranking generated by a WBFS. In this context, this thesis presents a study about the WBFS. First, a new voting approach is proposed to assess the stability of the WBFS, i.e, the variation in the result of the WBFS. Then, we evaluated the stability of the algorithms based on multilayer perceptron (MLP) and extreme learning machines (ELM). Furthermore, an improvement is proposed in the algorithms of Garson, Olden, and Yoon, combining them with the feature selector ReliefF. The new algorithms are called FSGR, FSOR, and FSYR. The experiments were performed based on 28 MLP architectures, 16 ELM architectures, and 16 data sets from the UCI Machine Learning Repository. The results show that there is a significant difference in WBFS stability depending on the training parameters of the ANNs and depending on the WBFS used. In addition, the proposed algorithms proved to be more effective than the classic algorithms. As far as we know, this study was the first attempt to measure the stability of WBFS, to investigate the effects of different ANN training parameters on the stability of WBFS, and the first to propose a combination of WBFS with another feature selector. Besides, the results provide information about the benefits and limitations of WBFS and represent a starting point for improving the stability of these algorithms. |
id |
UFG-2_2f1a2d669e1d2ca8cb7bff7713f19f98 |
---|---|
oai_identifier_str |
oai:repositorio.bc.ufg.br:tede/11396 |
network_acronym_str |
UFG-2 |
network_name_str |
Repositório Institucional da UFG |
repository_id_str |
|
spelling |
Barbosa, Rommel Melgaçohttp://lattes.cnpq.br/6228227125338610Barbosa, Rommel MelgaçoLima, Márcio Dias deLins, Isis DidierCosta, Ronaldo Martins daLeitão Júnior, Plínio de Sáhttp://lattes.cnpq.br/9968129748669015Costa, Nattane Luíza da2021-05-27T13:28:22Z2021-05-27T13:28:22Z2021-03-19COSTA, N. L. Uso e estabilidade de seletores de variáveis baseados nos pesos de conexão de redes neurais artificiais. 2021. 175 f. Tese (Doutorado em Ciência da Computação) - Universidade Federal de Goiás, Goiânia, 2021.http://repositorio.bc.ufg.br/tede/handle/tede/11396Artificial Neural Networks (ANN) are machine learning models used to solve problems in several research fields. Although, ANNs are often considered “black boxes”, which means that these models cannot be interpreted, as they do not provide explanatory information. Connection Weight Based Feature Selectors (WBFS) have been proposed to extract knowledge from ANNs. Most of studies that have been using these algorithms are based on just one ANN model. However, there are variations in the ANN connection weight values due to the initialization and training, and consequently, leading to variations in the importance ranking generated by a WBFS. In this context, this thesis presents a study about the WBFS. First, a new voting approach is proposed to assess the stability of the WBFS, i.e, the variation in the result of the WBFS. Then, we evaluated the stability of the algorithms based on multilayer perceptron (MLP) and extreme learning machines (ELM). Furthermore, an improvement is proposed in the algorithms of Garson, Olden, and Yoon, combining them with the feature selector ReliefF. The new algorithms are called FSGR, FSOR, and FSYR. The experiments were performed based on 28 MLP architectures, 16 ELM architectures, and 16 data sets from the UCI Machine Learning Repository. The results show that there is a significant difference in WBFS stability depending on the training parameters of the ANNs and depending on the WBFS used. In addition, the proposed algorithms proved to be more effective than the classic algorithms. As far as we know, this study was the first attempt to measure the stability of WBFS, to investigate the effects of different ANN training parameters on the stability of WBFS, and the first to propose a combination of WBFS with another feature selector. Besides, the results provide information about the benefits and limitations of WBFS and represent a starting point for improving the stability of these algorithms.Redes Neurais Artificiais (RNA) são modelos de aprendizado de máquina usados para a modelagem de problemas em vários campos de pesquisa. Contudo, as RNAs são frequentemente consideradas como “caixas pretas”, o que significa que esses modelos não podem ser interpretados, pois não fornecem informações explicativas. Os seletores de variáveis com base nos pesos de conexão (Weight Based feature selectors - WBFS) foram propostos para de extrair conhecimento das RNAs por meio de um ranking de importância das variáveis de entrada em relação à saída da rede. Grande parte das aplicações que utilizam esses algoritmos se baseiam em apenas um modelo de RNA. No entanto, existe uma variação nos valores dos pesos de conexão de RNAs treinadas em diferentes momentos de inicialização dos pesos e em razão do treinamento da rede, ocorrendo assim, uma variação no ranking de importância de um mesmo conjunto de dados. Neste contexto, esta tese apresenta um estudo sobre os WBFS. Primeiro, propomos uma nova abordagem de votação para avaliar a estabilidade dos WBFS, ou seja, a variação obtida ao gerar rankings de importância por meio de muitos modelos de RNA. Em seguida, avaliamos a estabilidade dos WBFS com base em multilayer perceptron (MLP) e extreme learning machines (ELM). Ademais, propomos uma melhoria nos WBFS de Garson, Olden e de Yoon, combinando-os com o seletor de variáveis ReliefF, denominados FSGR, FSOR e FSYR. Os experimentos foram realizados com base em 28 arquiteturas MLP, 16 arquiteturas ELM e 16 conjuntos de dados do repositório UCI. Os resultados mostram que há uma diferença significativa na estabilidade WBFS dependendo dos parâmetros de treinamento das RNAs e a depender do WBFS utilizado. Os algoritmos propostos se demonstraram mais eficazes do que os algoritmos clássicos. Até onde sabemos, este estudo foi a primeira tentativa de medir a estabilidade dos WBFS, de investigar os efeitos dos parâmetros da RNA na estabilidade do WBFS e o primeiro a propor uma combinação de WBFS com outro seletor de variáveis. Além disso, os resultados fornecem informações sobre os benefícios e as limitações dos WBFS e representam um ponto de partida para melhorar a estabilidade desses algoritmos.Submitted by Franciele Moreira (francielemoreyra@gmail.com) on 2021-05-26T18:11:30Z No. of bitstreams: 2 license_rdf: 805 bytes, checksum: 4460e5956bc1d1639be9ae6146a50347 (MD5) Tese - Nattane Luíza da Costa - 2021.pdf: 2785561 bytes, checksum: d29e283892397e024e286b14d240560d (MD5)Approved for entry into archive by Luciana Ferreira (lucgeral@gmail.com) on 2021-05-27T13:28:22Z (GMT) No. of bitstreams: 2 license_rdf: 805 bytes, checksum: 4460e5956bc1d1639be9ae6146a50347 (MD5) Tese - Nattane Luíza da Costa - 2021.pdf: 2785561 bytes, checksum: d29e283892397e024e286b14d240560d (MD5)Made available in DSpace on 2021-05-27T13:28:22Z (GMT). No. of bitstreams: 2 license_rdf: 805 bytes, checksum: 4460e5956bc1d1639be9ae6146a50347 (MD5) Tese - Nattane Luíza da Costa - 2021.pdf: 2785561 bytes, checksum: d29e283892397e024e286b14d240560d (MD5) Previous issue date: 2021-03-19porUniversidade Federal de GoiásPrograma de Pós-graduação em Ciência da Computação (INF)UFGBrasilInstituto de Informática - INF (RG)Attribution-NonCommercial-NoDerivatives 4.0 Internationalhttp://creativecommons.org/licenses/by-nc-nd/4.0/info:eu-repo/semantics/openAccessRedes neurais artificiaisSeleção de variáveisPesos de conexãoEstabilidadeRanking de importânciaArtificial neural networkFeature selectionConnection weightsStabilityCIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO::SISTEMAS DE COMPUTACAOUso e estabilidade de seletores de variáveis baseados nos pesos de conexão de redes neurais artificiaisThe use and stability of feature selectors based on artificial neural network connection weightsinfo:eu-repo/semantics/publishedVersioninfo:eu-repo/semantics/doctoralThesis2050050050026125reponame:Repositório Institucional da UFGinstname:Universidade Federal de Goiás (UFG)instacron:UFGLICENSElicense.txtlicense.txttext/plain; charset=utf-81748http://repositorio.bc.ufg.br/tede/bitstreams/3dc3986a-8fcf-4cf0-a7b9-65a4aa92d078/download8a4605be74aa9ea9d79846c1fba20a33MD51CC-LICENSElicense_rdflicense_rdfapplication/rdf+xml; charset=utf-8805http://repositorio.bc.ufg.br/tede/bitstreams/795fce0b-3fc3-48ed-9654-3c704463aefe/download4460e5956bc1d1639be9ae6146a50347MD52ORIGINALTese - Nattane Luíza da Costa - 2021.pdfTese - Nattane Luíza da Costa - 2021.pdfapplication/pdf2785561http://repositorio.bc.ufg.br/tede/bitstreams/48cadf3a-80c0-436d-90f8-159a215ee3ae/downloadd29e283892397e024e286b14d240560dMD53tede/113962024-01-09 10:01:04.653http://creativecommons.org/licenses/by-nc-nd/4.0/Attribution-NonCommercial-NoDerivatives 4.0 Internationalopen.accessoai:repositorio.bc.ufg.br:tede/11396http://repositorio.bc.ufg.br/tedeRepositório InstitucionalPUBhttp://repositorio.bc.ufg.br/oai/requesttasesdissertacoes.bc@ufg.bropendoar:2024-01-09T13:01:04Repositório Institucional da UFG - Universidade Federal de Goiás (UFG)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 |
dc.title.pt_BR.fl_str_mv |
Uso e estabilidade de seletores de variáveis baseados nos pesos de conexão de redes neurais artificiais |
dc.title.alternative.eng.fl_str_mv |
The use and stability of feature selectors based on artificial neural network connection weights |
title |
Uso e estabilidade de seletores de variáveis baseados nos pesos de conexão de redes neurais artificiais |
spellingShingle |
Uso e estabilidade de seletores de variáveis baseados nos pesos de conexão de redes neurais artificiais Costa, Nattane Luíza da Redes neurais artificiais Seleção de variáveis Pesos de conexão Estabilidade Ranking de importância Artificial neural network Feature selection Connection weights Stability CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO::SISTEMAS DE COMPUTACAO |
title_short |
Uso e estabilidade de seletores de variáveis baseados nos pesos de conexão de redes neurais artificiais |
title_full |
Uso e estabilidade de seletores de variáveis baseados nos pesos de conexão de redes neurais artificiais |
title_fullStr |
Uso e estabilidade de seletores de variáveis baseados nos pesos de conexão de redes neurais artificiais |
title_full_unstemmed |
Uso e estabilidade de seletores de variáveis baseados nos pesos de conexão de redes neurais artificiais |
title_sort |
Uso e estabilidade de seletores de variáveis baseados nos pesos de conexão de redes neurais artificiais |
author |
Costa, Nattane Luíza da |
author_facet |
Costa, Nattane Luíza da |
author_role |
author |
dc.contributor.advisor1.fl_str_mv |
Barbosa, Rommel Melgaço |
dc.contributor.advisor1Lattes.fl_str_mv |
http://lattes.cnpq.br/6228227125338610 |
dc.contributor.referee1.fl_str_mv |
Barbosa, Rommel Melgaço |
dc.contributor.referee2.fl_str_mv |
Lima, Márcio Dias de |
dc.contributor.referee3.fl_str_mv |
Lins, Isis Didier |
dc.contributor.referee4.fl_str_mv |
Costa, Ronaldo Martins da |
dc.contributor.referee5.fl_str_mv |
Leitão Júnior, Plínio de Sá |
dc.contributor.authorLattes.fl_str_mv |
http://lattes.cnpq.br/9968129748669015 |
dc.contributor.author.fl_str_mv |
Costa, Nattane Luíza da |
contributor_str_mv |
Barbosa, Rommel Melgaço Barbosa, Rommel Melgaço Lima, Márcio Dias de Lins, Isis Didier Costa, Ronaldo Martins da Leitão Júnior, Plínio de Sá |
dc.subject.por.fl_str_mv |
Redes neurais artificiais Seleção de variáveis Pesos de conexão Estabilidade Ranking de importância |
topic |
Redes neurais artificiais Seleção de variáveis Pesos de conexão Estabilidade Ranking de importância Artificial neural network Feature selection Connection weights Stability CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO::SISTEMAS DE COMPUTACAO |
dc.subject.eng.fl_str_mv |
Artificial neural network Feature selection Connection weights Stability |
dc.subject.cnpq.fl_str_mv |
CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO::SISTEMAS DE COMPUTACAO |
description |
Artificial Neural Networks (ANN) are machine learning models used to solve problems in several research fields. Although, ANNs are often considered “black boxes”, which means that these models cannot be interpreted, as they do not provide explanatory information. Connection Weight Based Feature Selectors (WBFS) have been proposed to extract knowledge from ANNs. Most of studies that have been using these algorithms are based on just one ANN model. However, there are variations in the ANN connection weight values due to the initialization and training, and consequently, leading to variations in the importance ranking generated by a WBFS. In this context, this thesis presents a study about the WBFS. First, a new voting approach is proposed to assess the stability of the WBFS, i.e, the variation in the result of the WBFS. Then, we evaluated the stability of the algorithms based on multilayer perceptron (MLP) and extreme learning machines (ELM). Furthermore, an improvement is proposed in the algorithms of Garson, Olden, and Yoon, combining them with the feature selector ReliefF. The new algorithms are called FSGR, FSOR, and FSYR. The experiments were performed based on 28 MLP architectures, 16 ELM architectures, and 16 data sets from the UCI Machine Learning Repository. The results show that there is a significant difference in WBFS stability depending on the training parameters of the ANNs and depending on the WBFS used. In addition, the proposed algorithms proved to be more effective than the classic algorithms. As far as we know, this study was the first attempt to measure the stability of WBFS, to investigate the effects of different ANN training parameters on the stability of WBFS, and the first to propose a combination of WBFS with another feature selector. Besides, the results provide information about the benefits and limitations of WBFS and represent a starting point for improving the stability of these algorithms. |
publishDate |
2021 |
dc.date.accessioned.fl_str_mv |
2021-05-27T13:28:22Z |
dc.date.available.fl_str_mv |
2021-05-27T13:28:22Z |
dc.date.issued.fl_str_mv |
2021-03-19 |
dc.type.status.fl_str_mv |
info:eu-repo/semantics/publishedVersion |
dc.type.driver.fl_str_mv |
info:eu-repo/semantics/doctoralThesis |
format |
doctoralThesis |
status_str |
publishedVersion |
dc.identifier.citation.fl_str_mv |
COSTA, N. L. Uso e estabilidade de seletores de variáveis baseados nos pesos de conexão de redes neurais artificiais. 2021. 175 f. Tese (Doutorado em Ciência da Computação) - Universidade Federal de Goiás, Goiânia, 2021. |
dc.identifier.uri.fl_str_mv |
http://repositorio.bc.ufg.br/tede/handle/tede/11396 |
identifier_str_mv |
COSTA, N. L. Uso e estabilidade de seletores de variáveis baseados nos pesos de conexão de redes neurais artificiais. 2021. 175 f. Tese (Doutorado em Ciência da Computação) - Universidade Federal de Goiás, Goiânia, 2021. |
url |
http://repositorio.bc.ufg.br/tede/handle/tede/11396 |
dc.language.iso.fl_str_mv |
por |
language |
por |
dc.relation.program.fl_str_mv |
20 |
dc.relation.confidence.fl_str_mv |
500 500 500 |
dc.relation.department.fl_str_mv |
26 |
dc.relation.cnpq.fl_str_mv |
125 |
dc.rights.driver.fl_str_mv |
Attribution-NonCommercial-NoDerivatives 4.0 International http://creativecommons.org/licenses/by-nc-nd/4.0/ info:eu-repo/semantics/openAccess |
rights_invalid_str_mv |
Attribution-NonCommercial-NoDerivatives 4.0 International http://creativecommons.org/licenses/by-nc-nd/4.0/ |
eu_rights_str_mv |
openAccess |
dc.publisher.none.fl_str_mv |
Universidade Federal de Goiás |
dc.publisher.program.fl_str_mv |
Programa de Pós-graduação em Ciência da Computação (INF) |
dc.publisher.initials.fl_str_mv |
UFG |
dc.publisher.country.fl_str_mv |
Brasil |
dc.publisher.department.fl_str_mv |
Instituto de Informática - INF (RG) |
publisher.none.fl_str_mv |
Universidade Federal de Goiás |
dc.source.none.fl_str_mv |
reponame:Repositório Institucional da UFG instname:Universidade Federal de Goiás (UFG) instacron:UFG |
instname_str |
Universidade Federal de Goiás (UFG) |
instacron_str |
UFG |
institution |
UFG |
reponame_str |
Repositório Institucional da UFG |
collection |
Repositório Institucional da UFG |
bitstream.url.fl_str_mv |
http://repositorio.bc.ufg.br/tede/bitstreams/3dc3986a-8fcf-4cf0-a7b9-65a4aa92d078/download http://repositorio.bc.ufg.br/tede/bitstreams/795fce0b-3fc3-48ed-9654-3c704463aefe/download http://repositorio.bc.ufg.br/tede/bitstreams/48cadf3a-80c0-436d-90f8-159a215ee3ae/download |
bitstream.checksum.fl_str_mv |
8a4605be74aa9ea9d79846c1fba20a33 4460e5956bc1d1639be9ae6146a50347 d29e283892397e024e286b14d240560d |
bitstream.checksumAlgorithm.fl_str_mv |
MD5 MD5 MD5 |
repository.name.fl_str_mv |
Repositório Institucional da UFG - Universidade Federal de Goiás (UFG) |
repository.mail.fl_str_mv |
tasesdissertacoes.bc@ufg.br |
_version_ |
1798044391788511232 |