Diamond Hoop: uma abordagem para otimização na composição de equipes no basquetebol com o uso de inteligência artificial sobre estatísticas pregressas

Abstract

Historically, it can be seen that it is undeniable that making the right decisions, or the best possible ones, with the resources available at the time, will always yield good results as a return. One area where this is crucial is in sports, which traditionally have a subjective bias. However, this has been changing in recent years thanks to efforts to apply quantitative methods and the evident results they bring. Among the existing problems, there is one that stands out as unsolved, which is the optimization in team composition. That is, forming a group that, when combined, possesses good synergy to generate positive outcomes as a product. This work specifically aims to seek a solution to this problem. To do so, a robust mechanism was employed that utilizes technologies within the field of Artificial Intelligence, leveraging a vast dataset from the NBA. The mechanism utilizes Graph Neural Networks to process the data and generate a vector representation that makes the problem representable. In order to achieve this, vectors of representation are extracted and used in a Results Simulator, which emulates a gaming environment between different teams and extracts the outcomes using a function designed for this purpose. Due to the number of simulations, they will be contained within the framework of the Law of Large Numbers, which, in a sense, makes this process analogous to a Monte Carlo Method. Thus, at the end of the process, the obtained response takes the form of a numerical approximation corresponding to various scenarios that could have unfolded. With this information at hand, several metrics were used to evaluate the similarity between distributions, which are already known in the literature, and clearly and objectively confirmed that the simulator achieved good adherence to reality.