Usando aprendizagem de máquina na criação de modelos para predizer resultados da Liga Nacional de Futsal do Brasil

Abstract

Based on the tremendous technological advances, the data volume has increased accordingly. Machine learning has gained relevance, helping many areas find patterns in large volumes of data. Sportisoneofthisarea. Researchesareusinglearningtechniquestopredictmatchresults and predict game situations. In this paper, we will present the steps to create ten models, based ontechniquesandconceptsofmachinelearning,topredicttheresultsofmatchesintheNational Futsal League of Brazil. The models use data from the first half of a match and historical factorsoftheteam. Duringthemodelingphase,featuresbasedonhistoricalfactorswerecreated that improved the performance of the model. The experiments performed show the models have good results concerning the accuracy, especially Random Forest and Gradient Boosting. Another contribution of this work is the committees of models. We use the committees to improve the overall accuracy, and we get better results about the individual performances. The predictions for the winning of the local team shows that the committees are essential in the prediction phase.