| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 5.18 MB | Adobe PDF |
Authors
Advisor(s)
Abstract(s)
Ao longo das últimas décadas, o basquetebol passou por uma evolução que transformou o que era um desporto para um negócio com enorme impacto social e financeiro. Essa transformação, aliada à constante necessidade de obter sucesso desportivo, criou a necessidade de inovação por parte de um clube desportivo de forma a distanciar-se dos seus adversários e de conquistar títulos e alcançar um maior lucro financeiro. A enorme quantidade de dados impossibilita a análise detalhada por um especialista e requer a utilização de meios computacionais para extrair informação valiosa. Alguns clubes já utilizam machine learning, mas este processo ainda se encontra numa fase inicial. Neste sentido, existe um enorme potencial para o tratamento e posterior valorização dos dados. A maioria das instituições desportivas identificaram a necessidade de capacidade técnica na análise de vastas fontes de informação, de forma que, as decisões tomadas sejam o mais fundamentadas possível, e, consequentemente, haja uma diminuição dos riscos na tomada das mesmas. Esta dissertação procura resolver este problema, e, com recurso ao mahine learning, mais concretamente do auxílio da metodologia CRISP-DM, passa pelo desenvolvimento de modelos de previsão de um rating de qualidade de um jogador e de equipa baseado em estatísticas de jogo, como por exemplo o número de pontos por jogo, com recurso a um modelo de previsão das mesmas. É também desenvolvido um modelo de previsão de resultados de jogos de basquetebol, tendo como base estatísticas de cada equipa, envolvendo diversas variáveis, de forma a tornar o modelo o mais robusto possível, e com uma maior flexibilidade.
Over the last few decades, basketball has undergone an evolution that has transformed what was a sport into a business with enormous social and financial impact. This transformation, combined with the constant need to achieve sporting success, created the need for innovation on the part of a sports club in order to distance itself from its opponents and win titles, and achieve greater financial profit. The huge amount of data makes detailed analysis impossible by an expert and requires the use of computational means to extract valuable information. Some clubs already use machine learning, but this process is still at an early stage. In this sense, there is an enormous potential for the processing and subsequent valorization of the data. Most sports institutions have identified the need for technical capacity in the analysis of vast sources of information so that the decisions taken are as well-founded as possible, and, consequently, there is a reduction in the risks in making them. This investigation seeks to solve this problem, and, using machine learning, more specifically with the help of the CRISP-DM methodology, it involves the development of a prediction model of a player and team quality rating based on game statistics, such as the number of points per game, using a prediction model. A model for predicting the results of basketball games will also be developed, based on statistics for each team, involving several variables, to make the model as robust as possible, and with greater flexibility.
Over the last few decades, basketball has undergone an evolution that has transformed what was a sport into a business with enormous social and financial impact. This transformation, combined with the constant need to achieve sporting success, created the need for innovation on the part of a sports club in order to distance itself from its opponents and win titles, and achieve greater financial profit. The huge amount of data makes detailed analysis impossible by an expert and requires the use of computational means to extract valuable information. Some clubs already use machine learning, but this process is still at an early stage. In this sense, there is an enormous potential for the processing and subsequent valorization of the data. Most sports institutions have identified the need for technical capacity in the analysis of vast sources of information so that the decisions taken are as well-founded as possible, and, consequently, there is a reduction in the risks in making them. This investigation seeks to solve this problem, and, using machine learning, more specifically with the help of the CRISP-DM methodology, it involves the development of a prediction model of a player and team quality rating based on game statistics, such as the number of points per game, using a prediction model. A model for predicting the results of basketball games will also be developed, based on statistics for each team, involving several variables, to make the model as robust as possible, and with greater flexibility.
Description
Keywords
Machine Learning CRISP-DM Rating Pontos Random forest Gradient Boosted Simple Regression Numeric Scorer Accuracy Previsão Points Prediction