Abstract
The college sports industry has grown tremendously over the past decade, with NCAA athletic departments recruiting almost half-a-million students to 19,866 teams in 2019 and generating $18.9 billion of revenue the same year. Sports analytics is one response to these growing needs, as its primary use in junior recruitment has presented fruitful for a litany of college sports programs across the nation. The technical aspect of this thesis portfolio pertains to improving sports analytics for the junior golf recruitment process while the STS research portion focuses on the social implications of similar algorithmic ranking.
GameForge, a golf analytics firm, aims to provide the same insights to college golf coaches by streamlining the recruitment of junior golfers to U.S. universities from around the world. GameForge seeks to develop a two-sided recruiting system that provides insights to junior players and their coaches as well as strengthen its predictive models with the inclusion of new data. For the capstone technical project, a systems-based approach was taken to develop data-driven machine learning models that would provide (a) a proprietary ranking system that compares junior athletes to one another; (b) a relative SWOT analysis that highlights each player’s strengths and skill gaps; and (c) a recommender system that suggests potential recruits to college coaches and recommends colleges of best fit to junior players.
The current approach in collegiate golf recruitment overlooks many golf players that have the potential to improve team performance. Top golfers are easily identified at tournaments and other major golfing events, but mid-level players are rarely considered due to the absence of a tangible platform to demonstrate their strengths. In addition to this, there is no current way for players to identify teams that are good matches based on metrics beyond rank, such as qualitative factors and personal preferences. This results in both colleges losing out on players that may strengthen their team and players not being able to find a team that will foster their skills and optimize their performance. The absence of a centralized setting that addresses the current recruitment concerns led GameForge to develop a data-driven platform.
The STS portion of the project researches the efficacy of machine learning ranking systems such as these and how to prevent bias that is often woven into artificial intelligence. Machine learning ranking systems are incredibly relevant in discussions on ethics in technology because large-scale ranking systems affect millions of people, and yet they are extremely susceptible to implicit and explicit algorithmic bias. At its core, artificial intelligence completes tasks efficiently, as exactly programmed and intended, without error. Unfortunately, if bias is ingrained in an algorithm, the issue will propagate from the algorithm to the product, and eventually to all the stakeholders. In the case of large-scale applications that utilize data to algorithmically determine “objective” rankings, the idea is noble because an unbiased ranking system will surely save time, money, and resources to produce a list of the best candidates. However, ranking systems may not be as effective as intended because once a candidate discovers the most important metrics in the algorithm, it is easy to take advantage of that knowledge. Another issue that must be considered is the frequency of historical bias embedded in the data that will appear in the algorithm results. If a manual process that is inherently biased becomes automated, it will still mirror its predecessor’s flaws. All things considered, the research will investigate these common problems through the lens of the Actor-Network Theory and the Social Construct of Technology. The goal of this research project is to be able to recognize the limitations of algorithmic systems and how to react with proper safety measures. The technical and research aspects of this project have clear benefits because these projects can be applied to other sports and their respective analytics while evaluating the efficacy of future algorithmic systems.
