Date of Award
Master of Science
Industrial and Manufacturing Systems Engineering
Sarah M. Ryan
Bettors on National Basketball Association (NBA) games commonly place wagers concerning the result of a game at time points during that game. We focus on the Totals (Over/Under) bet. Although many forecasting models have been built to predict the total number of points scored in an NBA game, they fail to provide bettors engaged in live-betting with predictions that are based on the game currently being played. We construct an Expert Bayesian Network to sequentially, as the game progresses, update the probability that the total points scored by both teams will exceed that set by the oddsmakers, and then use this probability to influence our wager at the end of the first, second, and third quarters. Research methods include data collection of team statistics over the last five NBA seasons, discretization of features, filter-based feature selection and specification of the network structure using domain knowledge and statistical tests. We compare the profit of our live-betting strategy against amateur betting strategies, wagers informed by a Naïve Bayes classifier, and wagers informed by a Bayesian Network whose structure is specified using a greedy search algorithm. When applied to games played during the early 2018-2019 NBA regular season, the Expert Bayesian Network and the Naïve Bayes model provide the most accurate predictions. Wagers informed by these two models yield profits of over 10% and 6%, respectively, but the other models and strategies are not profitable.
Enrique Marcos Alameda-Basora
Alameda-Basora, Enrique Marcos, "A dynamic Bayesian network to predict the total points scored in national basketball association games" (2019). Graduate Theses and Dissertations. 16955.