Predicting the winners of basketball games

Anže Kravanja (2013) Predicting the winners of basketball games. EngD thesis.

Preview

PDF Download (572Kb)

Abstract

The main goal of this thesis was to develop a prediction model for predicting basketball matches in the NBA. The problem was solved by using the techniques found in data mining, which include using machine learning algorithms, artificial intelligence and statistics. We present the whole path of development of such model from the initial parsing of so called ''play-by-play'' data, to the testing phase and choosing the final prediction model. We explored the qualitiy of attributes used, their connection to the class variable and help in prediction. While in testing phase, we tried different machine learning algorithms and we kept those whose predictions were promising. We combined the best algorithms in an ensemble to get even better results. We compared our results with betting odds, which give best predictions. At the end, we look deeper into the problems of our prediction model and gave suggestions about possible improvements and future work.

Actions (login required)