Activities
Date: Friday November 19th at 4 pm
Schedule-adjusted league tables during the football season
In this talk I will show how to construct a better football league table than the official ranking based on accumulated points to date. The aim of this work is (only) to produce a more informative representation of how teams currently stand, based on their match results to date in the current season; it is emphatically not about prediction. A more informative league table is one that takes proper account of "schedule strength" differences, i.e., differing numbers of matches played by each team (home and away), and differing current standings of the opponents that each team has faced.
This work extends previous "retrodictive" use of Bradley-Terry models and their generalizations, specifically to handle 3 points for a win, and also to incorporate home/away effects coherently without assuming homogeneity across teams. Playing records that are 100% or 0%, which can be problematic in standard Bradley-Terry approaches, are incorporated in a simple way without the need for a regularizing penalty on the likelihood. A maximum-entropy argument shows how the method developed here is the mathematically "best" way to account for schedule strength in a football league table.
Illustrations will be from the English Premier League.
The talk will be given by David Firth from the University of Warwick
DOUBLE EVENT! Date: Friday October 15th at 4 pm
Chaos in motion: the application and interpretation of entropy in biosignal analysis
The number of entropy statistics and studies employing them in the analysis of physiological time series has increased substantially over the last decade. In the context of sport science, the application of entropy remains relatively under-exploited despite the potential for entropy measures to quantify the complex nonlinear behaviour present in many real-world systems. Estimation of entropy typically involves several embedding parameters whose selection is critical in order to accurately quantify the underlying dynamics of the system under observation, with newer methods seeking to remove extraneous parameters altogether. However, the appropriate choice of embedding parameters has generally been overlooked, often leading researchers to infer a distorted meaning from entropy values and potentially draw misleading conclusions as a result.
This talk will describe entropy in the context of time series analysis, present an overview of the many entropy methods established in the scientific literature, and discuss the appropriate estimation of such measures for quantifying nonlinear behaviour in sport science applications with reference to a newly developed software toolkit called EntropyHub.
The talk will be given by Matthew Flood from the Luxembourg Institute of Health.
Application of Posterior predictive distribution estimators under the presence of additional information
In this talk, we present the application of the posterior predictive distribution when there exist some other sources of information available. To illustrate how well the proposed distribution estimators perform, we provide both a simulated study and real examples, by analyzing datasets of the National Hockey League (NHL), as well as soccer games.
The talk will be given by Abdolnasser Sadeghkani from the Instituto Tecnologico Autonomo de Mexico.
Date: Friday September 10th at 4 pm
A Hybrid Machine Learning Approach for the Modeling and Prediction of the UEFA EURO2020
Conventional approaches that analyze and predict the results of international matches in football are mostly based on the framework of Generalized Linear Models. The most frequently used type of regression models in the literature is the Poisson model. It has been shown that the predictive performance of such models can be improved by combining them with different regularization methods such as penalization.
More recently, also methods from the machine learning field such as boosting and random forests turned out to be very powerful in the prediction football match outcomes. Here, we analyze both a hybrid random forest extension based on conditional inference trees and a hybrid boosting extension based on extreme gradient boosting for modeling football matches. The models are fitted to match data from previous UEFA European Championships (EUROs) and based on the corresponding estimates all match outcomes of the EURO 2020 are repeatedly simulated (100,000 times), resulting in winning probabilities for all participating national teams.
The talk will be given by Andreas Groll from TU Dortmund University.
Date: Friday June 11th at 4 pm
To predict or protect; the value of screening in sports practice
Recently an academic discussion has evoked around the message that screening does not predict which athlete will sustain an injury. In various media the clinical implications of this conclusion are interpreted to mean that screening is useless in sports medical practice. However, screening remains essential in our efforts to protect athletes’ health. Indeed, screening may not predict which athlete will sustain an injury, but this does not directly make screening a useless tool in our tool bag. To extend what has been a robust discussion, this keynote will go in-depth into the concept of screening and explain some of the definitional issues that appear to fuel the heated debate. This address offers potential reasons why and how individual screening tests lack clinical utility, as well as what is needed to improve the value and efficiency of our instruments. This all leads up to the argument that screening can be important for to protect the health of an individual athlete, given some ground rules are needed and limitations exist.
The talk will be given by Evert Verhagen from the University of Amsterdam.
Date: Friday May 21st at 4 pm
Introducing Regularisation to Generalised Joint Regression Modelling and its Application to Football and Sports
When modelling the bivariate outcome of football matches and other sports, many different approaches regarding dependency have been investigated. We propose the use of copula regression via the powerful GJRM (Generalised Joint Regression Models) framework in R by Giampiero Marra and Rosalba Radice and present its use for modelling match results. Motivated by the application to football and FIFA World Cups in particular, we introduce two types of useful penalties. The first tackles a very specific issue occurring in sport tournaments and leagues (or other competitive situations), while the second is a Lasso-approximation yielding general sparsity.
The talk will be given by Hendrik van der Wurp from the Technical University of Dortmund.
Date: Friday April 23rd at 3.30 pm
Enduring Love: the Long-Term Effect of a New Stadium on Attendance at Professional English Soccer
Since 1988, 22 of the 92 of the english professional football clubs have acquired a new stadium. We estimate the causal effect of the new facility on attendance using a difference-in-difference model. We find that a new stadium raises attendance by around 20% on average, and this effect is sustained over nearly two decades. This result contrasts with the "Honeymoon effect" identified in the US literature, where attendance quickly reverts to the mean. We explain this by the promotion and relegation system. Unlike the closed major leagues, a new stadium represents an opportunity for a club in a lower division to generate increased revenue from fans, hire better players, win promotion and sustain a higher league position on average.
The talk will be given by Stefan Szymanski from the University of Michigan. The organization of this webinar is handled by James Reade from the University of Reading.
Date: Friday March 12th at 4 pm (or 3.30 pm)
Machine Learning in Orthopaedic Sports Medicine - Clinical Translation from the Registry to the Clinic
While the clinical application of machine learning to several health-care disciplines has increased considerably over recent years, it remains in its infancy in orthopaedic sports medicine. This presentation will review the basics of machine learning, give examples of how it has impacted other areas of orthopaedic surgery, and illustrate how machine learning can be applied to existing national knee ligament registers. Completed and ongoing collaborative studies between the University of Minnesota and the University of Oslo will be discussed, including a demonstration of how an in-clinic calculator was developed capable of estimating the risk of ACL reconstruction failure at a patient-specific level. Focusing on clinical translation of machine learning techniques, future opportunities within sports medicine will also be presented to stimulate idea generation and ultimately improve patient care.
R. Kyle Martin MD FRCSC is an orthopaedic sports medicine surgeon with the University of Minnesota. Originally from Canada, Dr. Martin completed his orthopaedic training at the University of Manitoba in 2017. He then travelled to Oslo, Norway where he spent one-year in a clinical fellowship with professor Lars Engebretsen at the University of Oslo and the Oslo Sports Trauma Research Center. Following the Oslo fellowship, Dr. Martin then completed a second clinical fellowship at Mayo Clinic in Rochester, MN. His current clinical practice revolves around knee, hip, and shoulder injuries and arthroscopy, while his research focus is on machine learning and its clinical applications to the field of sports medicine.
Date: Friday February 26th at 3.30 pm
Real-Time Skeleton Detection for Visual Sports Analysis and ... You
Dr. Manuel Stein presents a system for automated data acquisition and analysis from simple video recordings of team sport matches. The proposed system focuses on extracting movement data as well as body poses for players and allows tracking of ball movement in the case of ball-based forms of team sports. Furthermore, Dr. Manuel Stein will provide insights into his novel system for automatically displaying complex and advanced 2.5D visualizations superimposed on the original video recordings. As CEO & Co-Founder of Subsequent, he is especially interested in new project ideas and would like to hear about how you would like to make use of such a system.
Date: Friday January 22nd at 4pm
Shoe cushioning, body mass and running biomechanics as risk factors for running injury: A randomised trial with 800+ recreational runners
The objective of the event is to present a large randomised trial on the relationship between shoecushioning, running biomechanics and the risk of running-related injury to researchers interested ininjury aetiology, injury prediction and data sciences. The second part of the webinar will specificallyfocus on the challenges and opportunities resulting from the large dataset collected in this project, aswell as on recommendations for future research in injury prevention.
Introduction: Running-related injury research[1] – Daniel Theisen (ALAN)
The study design, data collection and descriptive data[2] – Laurent Malisoux (LIH)
Recently published results
a. Effect of shoe cushioning and body mass on injury risk[3]
b. Effect of cushioning on running biomechanics[4]
Let’s move to prediction
a. Predicting cumulative load using a wearable device[5] – Anne Backes (LIH)
b. Predicting running-related injury using machine learning – Hans van Eetvelde (UGent)
Date: Friday December 11th at 4pm
Statistical concept of CUB models to the world of sports
16:00 The class of CUB models: a paradigm for rating data (by Domenico Piccolo)
16:10 CUB models and extensions: from theory to action (by Rosaria Simone)
16:40: Focus on two developments: Nonlinear CUB and Treatment of "don't know" responses (by Marica Manisera)
16:55: Future research on CUB models in Sports: some insights (by Paola Zuccolotto)
If interested in joining the session, please contact Christophe Ley at Christophe.Ley@UGent.be
Date: Friday December 4th at 2.10pm
Prevention of injuries, are we heading the right direction?
Physical activity and sports are an integral part of our society. Both have a positive effect on quality of life. However, it should at the same time be noted that due to the physical demands the injury rate/percentage in sports are high[1]. Besides the consequences for the player, there are repercussions for the team and club. Injuries do not only lead to reduced performance, they cause financial losses as well[2]. In order to minimize these negative consequences, prevention programs to predict injuries were developed[3]. However, injuries show no tendency to decrease[4]. As such, it can be concluded that injury prevention is currently not sufficiently adequate and in need for change[5]. It is know that current approaches are not sufficiently addressing the complex and dynamic nature of sports injury aetiology, and this necessitates the need for integrating complex system approaches in sports injury prediction and prevention[6]. Accordingly, due to the lack of suitable methodological approaches, the use of Artificial intelligence to identify complex patterns of interactions and the implementation of wearables to continuously monitor the athletes have been introduced[6,7]. Despite the promising future for the use of AI and the implementation of wearables, further research, based on longitudinal studies with large datasets and continuous monitoring, is warranted to establish the effectiveness and predictive performance of these statistical techniques and methods in the particular domain of sports injury risk identification.
Speaker: Evi Wezenbeek (Ghent University)