In this thesis, the problem of predicting the elapsed time between two events in a basketball game is researched. First, our input data consisting of combinations of events in chronological order is studied, and then, the data is prepared and transformed so that it becomes better suited for machine learning. The elapsed time is predicted with the help of linear regression, regression trees and neural networks. For each algorithm, a short description is created; further, the best combination of independent variables and other parameters is found and finally, the best model is presented. To conclude, the best models of the utilized machine learning methods are compared. The best results were achieved by neural networks, while linear regression, as expected, proved to be the worst. Finally, the findings are presented and a few suggestions for improvements are added.
|
