How to Succeed in Basketball Without Really Trying: A support vector machine approach to draft picks and trade deals

Abstract

The National Basketball Association (NBA) is a men’s professional basketball league comprised of 30 teams and believed to be the highest level of play in the world. Teams regularly gain notoriety by accumulating game wins and championship rings. Many factors go into what makes a winning team, but the roster is arguably the most important aspect. This begs the question of possible optimal ways to choose what players will make up a team. This thesis will attempt to provide a strategic method of selection for both draft picks and trade deals when trying to reach a particular team goal. This method has been constructed to find the necessary player(s) using a two step process. First, the minimal change needed to reach a certain team goal must be calculated. Second, the player matching this calculation must be found. The minimal change will be found by using a support vector machine to calculate the hyperplane separating points classified as reaching the team goal, say beating the Cleveland Cavaliers, and falling short of the chosen team goal. The support vector machine will be built using training data comprised of wins, losses, and team statistics of a past, but recent, season. From this point, an integer program is used to find the players, whether draft picks, free agents, or both, necessary to move the data point to the “goal side” of the hyperplane. Using this, the player(s) that best fit(s) the team’s needs while minimizing cost can be found, and therefore make an optimal choice when it comes to trade deals and/or draft picks. Finally, this thesis will compare computed draft picks and trade deals with sample real world decisions and their later outcomes to test the validity of this approach.