Machine learning approaches to predict the match result: Brazilian futsal league case

• Autor(a) principal: Duarte, Denio
• Data de Publicação: 2021
• Outros Autores: Coppini, Jefferson Alexandre
• Tipo de documento: Artigo
• Idioma: por
• Título da fonte: Revista Brasileira de Futsal e Futebol
• Texto Completo: https://www.rbff.com.br/index.php/rbff/article/view/1110
• topic: Supervised Machine Learning; Prediction Models; Futsal; Aprendizado de Máquina Supervisionado; Modelo de Predição; Futsal

The use of machine learning approaches in sports has been grown in the last decade. Sports analytics, outcome match results, and possible player’s injury are examples of machine learning applications. Accordingly, this work aims to use machine learning techniques to build models to predict FutSal National League (LNF) results (win/loss/draw) based on data collected in the first half of a match. To accomplish that, we extract the data from the LNF website, and, based on the data, we propose six new features using the concept of team strength. The data correspond to the 2016 to 2019 seasons. The models are built usimg machine learning approaches, and they are validated through an accuracy metric. We build ten models, and the predictions are organized as follows: the individual performance of each model and a voting approach (committee) based on the majority of the predicted results. The results show that the individual models get better performance when predicting a single result (e.g., home win) with 95% accuracy. On the other hand, the committee gets a better performance regarding the overall results. The win, loss, and draw results reach almost 79% accuracy.