From computer screen to field: Students use sports analytics to predict performance outcomes

Students and faculty generate statistical models to predict ability and wins for the Cavalier football team, as well as other varsity and professional sports

sportsanalyrics

With new advances in technology, students and faculty researchers at the University have begun applying science to sports by using data analytics to predict the future success of both individual athletes and entire teams. William T. Scherer, professor of systems and information engineering and associate chair of Engineering Systems and Environment, along with engineering students completing their capstone projects, has been collaborating with the University’s football team for the past five years or so to optimize recruitment strategies and performance.

Additionally, students at HackCville — a local organization that offers semester-long courses in areas such as programming and data science — are also engaging in the field of sports analytics, studying outcomes of professional golf and basketball tournaments. Whether in University engineering courses or data science classes at HackCville, sports analytics continues to grow in popularity, especially with reports from the University football team that the information improves outcomes on and off the field.

One group of students advised by Scherer in recent years developed two models focused on recruiting. The first, the “Diamond in the Rough” model, predicts which lower-ranked high school football players might one day be in the NFL. Results can inform coaching staff about which recruits to pursue because, as Scherer said, outcompeting prominent football schools for high-profile recruits is unlikely. 

“There are some players out there with three or two [out of five] stars with incredible potential and we can get them,” Scherer said. “The model can predict whether they will have good forward success in college, and we found that there was actually very little correlation between composite scores provided and actual college performance.”

The second model returns a rating that attempts to measure an athlete’s grit, or how tough a player is. Though this characteristic of a player may appear difficult to measure empirically, Scherer and his students assigned the scores with the assistance of IBM Watson, a supercomputer that incorporates artificial intelligence and analytics. After IBM Watson attached certain personality traits to recruits based on their Twitter feeds, students used their own algorithm to determine the final scores.

“So we try to predict performance, but the other part of the equation is we want to pick who is going to fit well in the current U.Va. system, which is a hardcore, tough system with a rigorous coach that has high standards,” Scherer said.

Scherer and the engineering students recognize that football players are student athletes, meaning that their performance in the classroom can affect their ability to execute on the field too. As such, capstone projects aim to forecast the undergraduate GPA of possible recruits.

“We compute the likelihood that they will actually come to U.Va., how well we think they are going to perform, how tough they are or their grit and their GPA,” Scherer said. “For every high school player in America we can get the coaches’ estimates of those four things.”

Practices also serve as indicators of successful execution in games. Therefore, training content and exertion factor into overall performance. To give the Cavaliers a competitive edge, athletes on multiple teams, including football and field hockey, regularly wear sensors underneath their uniforms during practices and games. These “wearables” collect physiological data on heart rates and body positioning that capstone projects analyze to help direct the intensity and types of workouts players complete.  

Other research provides game-time recommendations. Scherer cited a shift in the football team’s mentality when it came to fourth downs, a change that is reflected across the sport. According to 10 years of data recorded for every play in Division I football, when facing the choice between fourth down and punting the ball, a team should play out the fourth down rather than punt the ball if the team is inside of their own 40 yard line.

“We have data on every player in college football in the last 15 years … and we can look at the statistics on what tends to work and what doesn’t work based on this kind of team you’re playing and the kind of offense you have,” Scherer said. 

Students seek opportunities to apply data analytics to sports outside of the University as well. Aaron Gu, third-year Engineering student and program leader of HackCville, has been investigating data released by the Professional Golfers' Association. Primarily, he is interested in a metric called strokes gained, which analyzes how well golfers are playing at specific distances from the green relative to professional PGA players. With this information, Gu can predict how well a player may perform in the next year, based on this year’s statistics.

As a program leader for HackCville, Gu also dedicates his time to teaching a course in data science. In this work, he instructs students on topics ranging from data analysis and visualization to machine learning. The program model encourages students to create projects centered on their own interests, which for some students means analyzing data from sports.

For example, one of his previous students examined the chances that NCAA basketball players had at becoming starters in the NBA. This model used machine learning to create an algorithm, which could assess current NCAA basketball players’ prospects based on previous data. It predicted that former college basketball players Kyle Guy and Zion Williamson both have high chances of starting.

“It's really [cool] seeing … our players at home,” Gu said, reflecting on his student’s project. “ We're having a good chance of being All-Stars in the NBA.”

While sports analytics in Charlottesville largely constitutes joint and independent faculty and student ventures in academia and HackCville at this time, Scherer revealed that one of the current engineering capstone projects proposes a plan for a new performance analytics center at the University. Since many coaches have expressed interest in incorporating data analytics into their programs, Scherer said, the students, in conjunction with the University’s athletic department, will outline an initiative that encompasses sports analytics research, and even potentially a sports analytics major and minor. The goal is to streamline the process of constructing and deploying predictive models for the majority, if not all, of the varsity sports teams at the University.

“We’re meeting with all the coaches of all the teams to see what they think, what are their needs, what would they want to do with the data,” Scherer said. “The fun thing about the performance analytics center is that you could develop the same kind of models for other sports, be it women’s golf, men’s lacrosse or swimming.”

related stories