| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 5.62 MB | Adobe PDF |
Authors
Advisor(s)
Abstract(s)
A popularidade global da NBA e o crescimento exponencial das apostas desportivas em
Portugal destacam a necessidade de modelos preditivos nesta área. Utilizando a metodologia
CRISP-DM, esta tese foca-se no desenvolvimento de um modelo eficaz para prever resultados
de jogos da NBA. O estudo identifica padrões cruciais para previsões precisas, analisando
dados históricos, estatísticas de jogadores e resultados de jogos.
As seis fases da metodologia incluem a compreensão dos objetivos do negócio, a exploração
de dados, a preparação meticulosa dos dados, a seleção do modelo, a avaliação rigorosa do
mesmo e a implementação final num site acessível aos utilizadores interessados. Na fase de
compreensão dos objetivos do negócio, foram definidos os requisitos e as metas do projeto.
Durante a exploração de dados, foram analisados e compreendidos os dados disponíveis. A
preparação dos dados envolveu a limpeza e transformação dos mesmos para garantir a sua
qualidade. Na fase de seleção do modelo, foram treinados diversos modelos, recorrendo a
vários algoritmos, com o objetivo de obter o melhor desempenho possível. A avaliação dos
modelos foi feita com base em várias métricas, sendo que o modelo escolhido atingiu uma
taxa de acerto de 64,4% e F1=72,4%. Finalmente, o modelo foi implementado num website
de fácil utilização, através da framework Streamlit .
A implementação num website aborda a atual falta de ferramentas eficazes de apoio à decisão
para prever jogos da NBA, contribuindo para a evolução do cenário de apostas desportivas
em Portugal.
The NBA’s global popularity and the exponential growth of sports betting in Portugal highlight the need for predictive models in this area. Using the CRISP-DM methodology, this thesis focuses on developing an effective model for forecasting NBA game outcomes. The study identifies patterns critical for precise predictions by analyzing historical data, player statistics, and game results. The six phases of the methodology include understanding business goals, data exploration, meticulous data preparation, model selection, its rigorous evaluation, and final deployment on an accessible website for the interested users. In the business objectives understanding phase, the project’s requirements and goals were defined. During data exploration, the available data was analyzed and insights about it were obtained. Data preparation involved cleaning and transforming the data to ensure its quality. In the model selection phase, multiple models were trained using various algorithms with the objective of obtaining the best possible performance. The models were evaluated based on multiple metrics, with the chosen model achieving an accuracy rate of 64.4% and F1=72,4%. Finally, the model was implemented on a user-friendly website using the Streamlit framework. The implementation on a website addresses the current lack of effective decision support tools for NBA game predictions, contributing to Portugal’s evolving sports betting landscape.
The NBA’s global popularity and the exponential growth of sports betting in Portugal highlight the need for predictive models in this area. Using the CRISP-DM methodology, this thesis focuses on developing an effective model for forecasting NBA game outcomes. The study identifies patterns critical for precise predictions by analyzing historical data, player statistics, and game results. The six phases of the methodology include understanding business goals, data exploration, meticulous data preparation, model selection, its rigorous evaluation, and final deployment on an accessible website for the interested users. In the business objectives understanding phase, the project’s requirements and goals were defined. During data exploration, the available data was analyzed and insights about it were obtained. Data preparation involved cleaning and transforming the data to ensure its quality. In the model selection phase, multiple models were trained using various algorithms with the objective of obtaining the best possible performance. The models were evaluated based on multiple metrics, with the chosen model achieving an accuracy rate of 64.4% and F1=72,4%. Finally, the model was implemented on a user-friendly website using the Streamlit framework. The implementation on a website addresses the current lack of effective decision support tools for NBA game predictions, contributing to Portugal’s evolving sports betting landscape.
Description
Keywords
NBA Sports betting Machine learning CRISP-DM Classification Apostas desportivas Classificação