The purpose of this study is apply STEAMS (Science, Technology, Engineering, Artificial Intelligence, Math, and Statistics) concepts to establish and to compare linear regression and neural network models in order to improve the prediction accuracy of NBA teams' winning records. Over one hundred basketball parameters of last ten years from 2010 to 2019 were collected from the NBA website and based on multivariate analysis of correlations and dependency, eighteen independent key parameters were chosen for the study. Both multivariate linear regression and neural network regression could predict NBA winning records accurately. However, the neural network model could further improve the prediction accuracy. These models can be utilized to interpret the winning strategies of each team. The number of hidden nodes for the artificial neural network was also optimized and it was found that one node is sufficient for this study without overfitting.
|