Αξιολόγηση ομάδων καλαθοσφαίρισης με τεχνικές μηχανικής μάθησης
Evaluating basketball teams with machine learning techniques
View/
Keywords
Sports analytics ; Machine learning ; Μηχανικη μαθηση ; K-means clustering ; Outcome predictionAbstract
Nowadays, the rapid development of machine learning is a useful tool in various
branches of everyday life. One of them is basketball.
In the context of the thesis, the relationship between basketball and machine
learning is explored. More specifically the factors that affect a team’s performance
will be studied, by identifying the modern playstyle through clustering. In addition,
the characteristics that give greater predictive ability are investigated, among classic
statistics and more analytical metrics. To achieve the above, data was collected from
the official NBA website and multiple machine learning methods were applied for
more accurate results.
Moreover, it is given an extensive look at the history of basketball, from the
earlier years to the introduction of analytics in basketball. An analytical literature
review is carried out and the theoretical background of the methods is additionally
investigated.
From the applications emerged the importance of players who are effective behind
the three-point line and are elite defenders, the fluidity of modern forwards and the
importance of big men. For predictive models it was found that the best method
differs depending on the choice of features. Convolutional neural networks and
logistic regression gave the best performance for different sets of variables.