Chapter 8 Conclusion
Duke Men’s Basketball has a vast and rich dataset that has much to be explored. Of particular interest is how a player interacts against his teammates and defenders. This paper focuses on modeling player interactions via passing networks–network centrality and betweenness scores identify key players within a team. By evaluating passing networks, not only can a player’s value within a team be deduced, but also how a player’s value within a team has changed over time. Modeling each posession in a game with network characteristics as features can be directionally useful. A more robust approach utilizes Peter Hoff’s AMEN package, which models both nodal and network characteristics. The results through this approach similarly show the significance of passing and receiving the ball. Teamwork and high collaboration leads to successful plays.
8.1 Future Steps
The scope of this work captures possessions of a game on an individual level. However, using the output of the latent eigenvalue model to predict nodal attributes is not sufficient to capture the game fully. Currently, an implementation and adaptation of a model influenced by Luke Bornn’s “A Multiresolution Stochastic Process Model for Predicting Basketball” aims to capture the game of basketball more robustly. Future work will use this advanced model to create metrics for assessing a player’s production value on a team level. A summary of the model replication can be found in Appendix A.