Um arcabouço computacional para caracterizar sorte e habilidade em ligas esportivas

• Ano de defesa: 2017
• Autor(a) principal: Raquel Yuri da Silveira Aoki
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Minas Gerais; UFMG
• Palavras-chave em Português: Estatística Bayesiana; Análise Esportiva; Modelos Gráficos Probabilísticos; Computação; Análise esportiva; Teoria bayesiana de decisão estatistica; Modelos gráficos probabilísticos
• Link de acesso: http://hdl.handle.net/1843/ESBF-ANZK4X

It is well know that predicting the outcome of sports events from feature-based models is a very challenging task. This work quantifies that difficulty by proposing a coefficient that measures the impact of luck and skill on final results of sport leagues. It was collected and analyzed all games from 1503 seasons of 84 countries in 4 different sports: basketball, soccer, volleyball and handball. Though the proposed coefficient, it was possible to measure the competitiveness by countries and sports. Another contribution of this work is the fitting of a probabilistic graphical model in order to learn about the teams' skill in situations where skill has a large impact on the results according with the proposed coefficient. The model finds on National Basketball Association (NBA) championship a correlation of about 0.7 between the estimated skill and the number of wins throughout the season.