COSTA, Í. B.; http://lattes.cnpq.br/2851866994953797; COSTA, Ígor Barbosa da.
Resumen:
In soccer, predicting soccer results is a challenging task due to the sport’s stochastic nature and the complexity of the innumerable factors that influence the match’s result. In
the last decade, with the evolution in the techniques of acquisition, storage, and processing of large volumes of data, several online data sources (e.g., textual and tabular) appeared. These sources contain information on played matches (e.g., lineups, scores, and scouts) and information on betting market movements. In such a scenario, data mining and machine learning techniques can be a viable alternative for discovering patterns in historical data and, consequently, for the prediction of results. This work explores different machine learning techniques, ranging from simple classifiers to complex neural networks, to make predictions both before the start of matches and during the match (in real-time), adjusting the probabilities of results as soon as new events occur (e.g., goals and red cards). In the case of pre-game prediction, this work addresses a problem that is still little explored in the literature: "will both teams score goals?". This type of prediction has aroused growing interest in recent years, due to the growth of the sports betting market. In the case of prediction during the match, the objective is to predict the final result of the game. A set of experiments was carried out to evaluate different techniques in terms of accuracy and profitability in the betting market. The results obtained are important in several aspects, ranging from evaluating the factors that influence the outcome of a match to the analysis of the efficiency of machine learning techniques in the sports betting market.