Publicação
Techniques to reject atypical patterns
| Resumo: | Supervised Classification algorithms are only trained to recognize and classify certain patterns, those contained in the training group. Therefore, these will by default, classify the unknown patterns incorrectly, causing unwanted results. This work proposes several solutions, to make the referred algorithms capable of detecting unknown patterns. The main approach for the development of models capable of recognizing these patterns, was the use of three different models of Autoencoders: Simple Autoencoder (SAE), Convolutional Autoencoder (CAE) and Variational Autoencoder (VAE), that are a specific type of Neural Networks. After carrying out several tests on each of the three models of Autoencoders, it was possible to determine which one performed best the task of detecting/rejecting atypical patterns. Afterwards, the performance of the best Autoencoder was compared to the performance of a Convolutional Neural Network (CNN) in the execution of the referred task. The conclusion was that the VAE effectively detected atypical patterns better than the CNN. Some conventional Machine Learning techniques (Support Vector Machine (SVM), Random Forest (RF), Logistic Regression (LR)) were also tested. The one that presented the best performance was the RF classifier, achieving an accuracy of 75% in the detection of atypical/typical patterns. Thus, regarding the classification balance between atypical and typical patterns, Machine Learning techniques were not enough to surpass the Deep Learning methods, where the best accuracy reached 88% for the VAE. |
|---|---|
| Autores principais: | Lopes, Júlio Castro |
| Outros Autores: | Rodrigues, Pedro João |
| Assunto: | Deep learning |
| Ano: | 2022 |
| País: | Portugal |
| Tipo de documento: | comunicação em conferência |
| Tipo de acesso: | acesso restrito |
| Instituição associada: | Instituto Politécnico de Bragança |
| Idioma: | inglês |
| Origem: | Biblioteca Digital do IPB |
| _version_ | 1867173196549062656 |
|---|---|
| author | Lopes, Júlio Castro |
| author2 | Rodrigues, Pedro João |
| author2_role | author |
| author_facet | Lopes, Júlio Castro Rodrigues, Pedro João |
| author_role | author |
| contributor_name_str_mv | Biblioteca Digital do IPB |
| country_str | PT |
| creators_json_txt | [{\"Person.name\":\"Lopes, Júlio Castro\"},{\"Person.name\":\"Rodrigues, Pedro João\",\"Person.identifier.orcid\":\"0000-0002-0555-2029\"}] |
| datacite.contributors.contributor.contributorName.fl_str_mv | Biblioteca Digital do IPB |
| datacite.creators.creator.creatorName.fl_str_mv | Lopes, Júlio Castro Rodrigues, Pedro João |
| datacite.date.Accepted.fl_str_mv | 2022-01-01T00:00:00Z |
| datacite.date.available.fl_str_mv | 2023-03-10T14:48:57Z |
| datacite.date.embargoed.fl_str_mv | 2023-03-10T14:48:57Z |
| datacite.rights.fl_str_mv | http://purl.org/coar/access_right/c_16ec |
| datacite.subjects.subject.fl_str_mv | Deep learning |
| datacite.titles.title.fl_str_mv | Techniques to reject atypical patterns |
| dc.contributor.none.fl_str_mv | Biblioteca Digital do IPB |
| dc.creator.none.fl_str_mv | Lopes, Júlio Castro Rodrigues, Pedro João |
| dc.date.Accepted.fl_str_mv | 2022-01-01T00:00:00Z |
| dc.date.available.fl_str_mv | 2023-03-10T14:48:57Z |
| dc.date.embargoed.fl_str_mv | 2023-03-10T14:48:57Z |
| dc.format.none.fl_str_mv | application/pdf |
| dc.identifier.none.fl_str_mv | http://hdl.handle.net/10198/27622 |
| dc.language.none.fl_str_mv | eng |
| dc.publisher.none.fl_str_mv | Springer |
| dc.rights.cclincense.fl_str_mv | http://creativecommons.org/licenses/by/4.0/ |
| dc.rights.none.fl_str_mv | http://purl.org/coar/access_right/c_16ec |
| dc.subject.none.fl_str_mv | Deep learning |
| dc.title.fl_str_mv | Techniques to reject atypical patterns |
| dc.type.none.fl_str_mv | http://purl.org/coar/resource_type/c_5794 |
| description | Supervised Classification algorithms are only trained to recognize and classify certain patterns, those contained in the training group. Therefore, these will by default, classify the unknown patterns incorrectly, causing unwanted results. This work proposes several solutions, to make the referred algorithms capable of detecting unknown patterns. The main approach for the development of models capable of recognizing these patterns, was the use of three different models of Autoencoders: Simple Autoencoder (SAE), Convolutional Autoencoder (CAE) and Variational Autoencoder (VAE), that are a specific type of Neural Networks. After carrying out several tests on each of the three models of Autoencoders, it was possible to determine which one performed best the task of detecting/rejecting atypical patterns. Afterwards, the performance of the best Autoencoder was compared to the performance of a Convolutional Neural Network (CNN) in the execution of the referred task. The conclusion was that the VAE effectively detected atypical patterns better than the CNN. Some conventional Machine Learning techniques (Support Vector Machine (SVM), Random Forest (RF), Logistic Regression (LR)) were also tested. The one that presented the best performance was the RF classifier, achieving an accuracy of 75% in the detection of atypical/typical patterns. Thus, regarding the classification balance between atypical and typical patterns, Machine Learning techniques were not enough to surpass the Deep Learning methods, where the best accuracy reached 88% for the VAE. |
| dirty | 0 |
| eu_rights_str_mv | restrictedAccess |
| format | conferencePaper |
| fulltext.url.fl_str_mv | https://bibliotecadigital.ipb.pt/bitstreams/78f3c55d-d04a-42bd-ae51-7d7c8b005bbc/download |
| id | ipb_ca80f2b5f893ddbf7ebec6e1fd7bd151 |
| identifier.url.fl_str_mv | http://hdl.handle.net/10198/27622 |
| instacron_str | ipb |
| institution | Instituto Politécnico de Bragança |
| instname_str | Instituto Politécnico de Bragança |
| language | eng |
| network_acronym_str | ipb |
| network_name_str | Biblioteca Digital do IPB |
| oai_identifier_str | oai:bibliotecadigital.ipb.pt:10198/27622 |
| organization_str_mv | urn:organizationAcronym:ipb |
| person_str_mv | Lopes, Júlio Castro Rodrigues, Pedro João Rodrigues, Pedro João https://www.ciencia-id.pt/1316-21BB-9015 1316-21BB-9015 http://orcid.org/0000-0002-0555-2029 0000-0002-0555-2029 |
| publishDate | 2022 |
| publisher.none.fl_str_mv | Springer |
| reponame_str | Biblioteca Digital do IPB |
| repository_id_str | urn:repositoryAcronym:ipb |
| service_str_mv | urn:repositoryAcronym:ipb |
| spelling | engSpringerpt_PTSupervised Classification algorithms are only trained to recognize and classify certain patterns, those contained in the training group. Therefore, these will by default, classify the unknown patterns incorrectly, causing unwanted results. This work proposes several solutions, to make the referred algorithms capable of detecting unknown patterns. The main approach for the development of models capable of recognizing these patterns, was the use of three different models of Autoencoders: Simple Autoencoder (SAE), Convolutional Autoencoder (CAE) and Variational Autoencoder (VAE), that are a specific type of Neural Networks. After carrying out several tests on each of the three models of Autoencoders, it was possible to determine which one performed best the task of detecting/rejecting atypical patterns. Afterwards, the performance of the best Autoencoder was compared to the performance of a Convolutional Neural Network (CNN) in the execution of the referred task. The conclusion was that the VAE effectively detected atypical patterns better than the CNN. Some conventional Machine Learning techniques (Support Vector Machine (SVM), Random Forest (RF), Logistic Regression (LR)) were also tested. The one that presented the best performance was the RF classifier, achieving an accuracy of 75% in the detection of atypical/typical patterns. Thus, regarding the classification balance between atypical and typical patterns, Machine Learning techniques were not enough to surpass the Deep Learning methods, where the best accuracy reached 88% for the VAE.application/pdfpt_PTTechniques to reject atypical patternsLopes, Júlio CastroPersonalRodrigues, Pedro JoãoDSpacehttp://dspace.org/items/6c5911a6-b62b-4876-9def-60096b52383aDSpacehttp://dspace.org/items/6c5911a6-b62b-4876-9def-60096b52383aRodriguesPedro JoãoCiência IDhttps://www.ciencia-id.pt1316-21BB-9015ORCIDhttp://orcid.org0000-0002-0555-2029HostingInstitutionOrganizationalBiblioteca Digital do IPBe-mailmailto:dspace@ipb.ptdspace@ipb.ptISBNIsPartOf978-3-031-23236-7DOIIsPartOfhttps://doi.org/10.1007/978-3-031-23236-7_12023-03-10T14:48:57Z20222022-01-01T00:00:00ZHandlehttp://hdl.handle.net/10198/27622http://purl.org/coar/access_right/c_16ecrestricted accessDeep learning311736 bytesother research producthttp://purl.org/coar/resource_type/c_5794conference paper2022http://creativecommons.org/licenses/by/4.0/http://purl.org/coar/access_right/c_16ecapplication/pdffulltexthttps://bibliotecadigital.ipb.pt/bitstreams/78f3c55d-d04a-42bd-ae51-7d7c8b005bbc/downloadInternational Conference on Optimization, Learning Algorithms and Applications - OL2A 20221754318Bragança |
| spellingShingle | Techniques to reject atypical patterns Lopes, Júlio Castro Deep learning |
| status | SINGLETON |
| subject.fl_str_mv | Deep learning |
| title | Techniques to reject atypical patterns |
| title_full | Techniques to reject atypical patterns |
| title_fullStr | Techniques to reject atypical patterns |
| title_full_unstemmed | Techniques to reject atypical patterns |
| title_short | Techniques to reject atypical patterns |
| title_sort | Techniques to reject atypical patterns |
| topic | Deep learning |
| topic_facet | Deep learning |
| url | http://hdl.handle.net/10198/27622 |
| visible | 1 |