Acquisition and analysis of data from futsal matches

Abstract

The domain of this master’s thesis is futsal. More precisely, this thesis has been elaborated in collaboration wit the futsal team of Girona: Girona Escola de Futbol Sala (GEFS). The idea of this club is to grow as much it is possible in order to become a futsal reference. However, it must be taken into account that the growth of a sports club is closely linked to the performance of the matches. In fact, this is where the main motivation and justification for this master’s thesis lies. Each weekend, GEFS collects two sheets of paper with data about the match (papers M and E). The club collects all this data because it wants to make datadriven decisions regarding the performances of the players. So, for this thesis there have been defined these three aims: • Digitalize existing data and propose new ways of recording match information to facilitate automatic extraction in the future. • Structure the data for its maintenance, management, and analysis. • Use the data to better understand the matches, investigate the performance of each player, improve collective and group performance and facilitate the preparation of the following training sessions. To achieve them the results have been divided into two groups. On the one hand, the non-visual results are related to the first two aims. It has been proposed an automatic image process to extract the data directly from a photo. In other words, the data that GEFS collects by hand in two sheets of paper can be digitalized with only taking one photo. Also, it has been proposed a new way to collect the data in order to improve this process. Furthermore, it has been designed a relational database that stores all the collected data in a coherent way. It is divided into static and growing datasets and it is located at the Google Cloud Platform. Every time a match is played, it has been created a pipeline that digitalize the data and, afterwards, it appends it to the correspondent table from the database. On the other hand, the visual results are linked to the third aim. There are three outputs. First, it has been produced an interactive dashboard with the requested graphs by the coach of GEFS. Secondly, it has been used the Principal Component Analysis to determine that the features Lost, Tackle, TG_received and TG_scored are the most relevant to discriminate between the performances of ii the players. Lastly, to estimate the density function and to cluster the players there have been used Gaussian Mixtures. The adjusted model is an EEE (ellipsoidal, equal volume, shape, and orientation) with 7 components. Consequently, there have been created seven clusters. Each one had their own logical interpretation