Loading...
Publication
Sistema de recomendação híbrido para definição de táticas de futebol 11
| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 4.11 MB | Adobe PDF |
Advisor(s)
Abstract(s)
Esta dissertação aborda a definição de táticas no futebol de 11, um desafio complexo especialmente
em equipas amadoras, onde a falta de recursos tecnológicos limita a capacidade
de análise aprofundada de dados de jogo. O objetivo central consistiu em desenvolver e demonstrar
um sistema de recomendação híbrido que auxilie treinadores na tomada de decisões
táticas, integrando dados de desempenho da própria equipa e dos adversários.
Para concretizar este objetivo, foi implementada uma aplicação web composta por um backend
robusto em Java, um módulo de recomendação baseado em técnicas de machine
learning (Random Forest) e uma interface em Angular. A metodologia seguiu as diretrizes
PRISMA para revisão sistemática da literatura, analisando boas práticas na área dos sistemas
de recomendação e da análise de dados desportivos. O sistema conjuga filtragem
baseada em conteúdo com elementos inspirados na filtragem colaborativa, permitindo gerar
sugestões personalizadas de formações e estratégias.
Os testes realizados incluíram a validação do módulo de recomendação com dados reais e
a avaliação técnica da aplicação, evidenciando o seu funcionamento adequado e o potencial
para apoiar decisões fundamentadas. Embora não tenha sido possível recolher respostas
diretas de treinadores, foi preparado um questionário de avaliação para recolher futuros
contributos de treinadores de formação do Porto, cujas respostas poderão orientar melhorias
incrementais.
Conclui-se que a solução proposta representa um passo viável na democratização do uso de
inteligência artificial no futebol, oferecendo uma ferramenta de apoio acessível e aplicável
em contextos de equipas de formação, onde o impacto de decisões mais informadas poderá
ser particularmente relevante para a evolução técnica e tática das equipas.
This dissertation addresses the challenge of defining tactics in eleven-a-side football, a particularly complex task for amateur teams where the lack of technological resources limits the capacity for in-depth match data analysis. The main objective was to develop and demonstrate a hybrid recommendation system to support coaches in tactical decision-making, integrating performance data from both their own team and opponents. To achieve this goal, a web application was implemented, comprising a robust Java-based backend, a recommendation module based on machine learning techniques (Random Forest), and a frontend in Angular. The methodology followed PRISMA guidelines for systematic literature review, analyzing best practices in the field of recommendation systems and sports data analysis. The system combines content-based filtering with elements inspired by collaborative filtering, enabling the generation of personalized suggestions for formations and strategies. The tests included the validation of the recommendation module with real data and the technical evaluation of the application, demonstrating its correct operation and its potential to support evidence-based decisions. Although it was not possible to collect direct feedback from coaches, an evaluation questionnaire was prepared to gather future input from youth coaches in Porto, whose responses could guide incremental improvements. In conclusion, the proposed solution represents a viable step toward democratizing the use of artificial intelligence in football, providing an accessible decision-support tool that can be applied in youth football contexts, where the impact of more informed tactical decisions could be particularly relevant for the technical and strategic evolution of teams.
This dissertation addresses the challenge of defining tactics in eleven-a-side football, a particularly complex task for amateur teams where the lack of technological resources limits the capacity for in-depth match data analysis. The main objective was to develop and demonstrate a hybrid recommendation system to support coaches in tactical decision-making, integrating performance data from both their own team and opponents. To achieve this goal, a web application was implemented, comprising a robust Java-based backend, a recommendation module based on machine learning techniques (Random Forest), and a frontend in Angular. The methodology followed PRISMA guidelines for systematic literature review, analyzing best practices in the field of recommendation systems and sports data analysis. The system combines content-based filtering with elements inspired by collaborative filtering, enabling the generation of personalized suggestions for formations and strategies. The tests included the validation of the recommendation module with real data and the technical evaluation of the application, demonstrating its correct operation and its potential to support evidence-based decisions. Although it was not possible to collect direct feedback from coaches, an evaluation questionnaire was prepared to gather future input from youth coaches in Porto, whose responses could guide incremental improvements. In conclusion, the proposed solution represents a viable step toward democratizing the use of artificial intelligence in football, providing an accessible decision-support tool that can be applied in youth football contexts, where the impact of more informed tactical decisions could be particularly relevant for the technical and strategic evolution of teams.
Description
Keywords
Sistemas de Recomendação Machine Learning Futebol de 11 tomada de decisão tática equipas de formação Decision making Recommendation Systems Tactical decision-making 11-a-side football