Modeling and Predicting the Outcomes of NBA Basketball Games

Point spreads betting based on the approximation of score difference is prevalent within the NBA community. In this work, the primary objective is to construct a multidimensional linear model that incorporates both game and player information to forecast the score difference of NBA basketball games. We first considered the SHT Model that contains merely game information, including team intrinsic Strength, Home court advantage, and Tiredness due to successive games. The model is adequately convincing since the estimates are legitimate in terms of their signs and magnitudes. We further examined the influence of the absence of players on forecasting the score difference and applied the respective salary as a weighting factor to add variability into the model. The prediction results indicate that the SHT Model has the highest accuracy in predicting score difference, which will be adopted as our final model. Tests on different seasons demonstrate the final model's stability and practice ability.