Aplicação de técnicas de mineração de dados para a identificação de padrões em partidas de futebol

Abstract

The growth in the collection and storage of football data, driven by sensors, tracking software, and analytical platforms, has expanded the possibilities for applying Data Science in sports. In this context, Data Mining stands out as a promising approach to explore patterns and understand relationships between technical attributes and team behavior. This study aimed to identify patterns in football matches by applying clustering algorithms to a public dataset containing match statistics. First, different datasets were compared, and one was selected with information on match events (goals, shots, possession, corners, crosses, fouls, and cards). An extract, transform and load process was then carried out to normalize the data and validate record consistency. Next, performance vectors were constructed using temporal match windows, combining five-match histories with aggregated statistics (means and standard deviations over 3, 4, and 5 match windows), which served as input to unsupervised clustering algorithms. The results indicate that the resulting clusters primarily capture differences in style and intensity across leagues, reflecting recent performance patterns, but show low direct correspondence with match outcomes (win, draw, or loss). Therefore, these findings support the use of Data Mining techniques as an exploratory tool in football analytics, providing a basis for future studies in Sports Data Science.