Utilização do classificador polinomial como ferramenta de predição de resultados de partidas de futebol

• Ano de defesa: 2017
• Autor(a) principal: Martins, Rodrigo Grassi
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Uberlândia; Brasil; Programa de Pós-graduação em Engenharia Elétrica
• Palavras-chave em Português: Engenharia elétrica; Reconhecimento de padrões; Esportes coletivos; Futebol; Classificador polinomial; Predição de partidas de futebol; Seleção de características; Polynomial classifier; Recognition of patterns; Prediction of football matches,; Selection of features.; CNPQ::ENGENHARIAS::ENGENHARIA ELETRICA
• Link de acesso: https://repositorio.ufu.br/handle/123456789/21365; http://doi.org/10.14393/ufu.te.2017.171

The interest of so many people in the world for football generates not only viewers but also many nancial movements around football. Computer systems that work with predic- ting results and help minimize risk and maximize pro ts make it an important tool for the day-to-day running of football. The working hypothesis of this thesis is that the Polynomial Classi er, a technique widely used as a standard classi er, can also be used as a feature selection algorithm. In order to investigate the prediction of the results of soccer matches. The Naive Bayes, decision tree, MLP, RBF and SVM algorithms were chosen. The choice of these classi ers is based on the state of the art. To validate the e cacy of the polynomial classi er, tests were carried out using the following methods: Principal Component Analy- sis(PCA) and Relief. The data used for the proposed approach were the results of soccer matches, obtained in the following championships: English championship season 2014/15 (CI 2014/15), Spanish championship season 2014/2015 (CE 2014/15) and Brazilian cham- pionships seasons of 2010 (CB 2010) and 2012 (CB 2012). The validation techniques used were: cross validation and sliding window. The results obtained according to the proposed methodology show that the CP obtained the best accuracy when compared to the other ve classi ers used: Naive Bayes(NB), Decision Tree(DT), Multilayer Perceptron(MLP), Radial Basis Function(RBF) e Supportt Vector Machine(SVM). It is still possible to a rm that from agreement with the results the CP was able to improve the accuracy of the ve clas- si ers with indices higher than Relief and PCA. It is also possible to state according to the results presented in section ref ArtData the accuracy obtained with this methodology is as good or superior to the results found in the state of the art, varying from 0.96 to 0.99. Keywords: Polynomial classi er, Recognition of patterns, prediction of football matches, selection of features.