The Seminar Series Presents:
Kostas Pelechrinis, PhD
Professor Pelechrinis is an Associate Professor at the School of Computing and Information at the University of Pittsburgh.He holds a PhD degree in Computer Science from the University of California, Riverside and his research interests include network and data science, applied machine learning with an emphasis on applications in urban informatics and sports analytics. He is a recipient of the prestigious Young Investigator Award from the Army Research Office for his work on heterogeneous networks, while his research is also supported from the National Science Foundation. He has collaborated with professional sports teams.
Dr. Pelechrinis will present "Soccer analytics: Past, Present, and Future" on Friday, October 5, 2018 in 1811 Posvar Hall.
Abstract: Soccer is one of the least quantified sports to date. One of the main reasons is the fact that most statistics to date capture on-ball events, while soccer is a game of space and off-ball movement. In this talk, I will start by presenting a brief history of soccer analytics, and the most popular advanced metric to date, i.e., expected goals (xG). While xG has been used to describe the performance of a team there is still not a good way to explicitly quantify the contribution of every player on the field to his team chances of winning. For example, successful advanced metrics such as the (adjusted) +/- that allows for division of credit among a basketball team's players, fail to work in soccer due to severe co-linearities. Using data from (i) approximately 20,000 games from 11 European leagues for 8 seasons, as well as, (ii) player ratings from FIFA, I will present a Skellam regression model that can estimate the importance of every line in winning a soccer game. This model can then be translated to expected league points added (per game) above a replacement player method and consequently be used as a guide for contracts' monetary value decisions. For example, using market value data for approximately 10,000 players we further identify that currently the market clearly under-values defensive line players relative to goalkeepers. Finally, we discuss how this model can be significantly enhanced using optical tracking data, but also, how it can be used to obtain positional (and consequently player) value for American football (another sports where achieving division of credit has been proven to be hard to date).
Location and Address
1811 Welsey W. Posvar Hall
230 S. Bouquet St.
Pittsburgh, PA 15260