The sports industry’s leading data scientists make up Stats Perform’s world-class AI team. Find out how Stats Perform continues revolutionising sports through technological advancements.
The Stats Perform AI Team leverage advancements in artificial intelligence and machine learning to develop new levels of sports innovation.
From enhancing business intelligence dashboards and internal performance reports to targeted content on an unprecedented scale, Automated Insights helps organisations make better decisions through the power of artificial intelligence and data-driven stories.
Through a self-service platform that enables complete narrative customisation, real-time content updates, and a powerful API for flexible publishing.
The AI in Sport Series began by providing members of the sports industry with a solid grounding in the role Artificial Intelligence plays in maximizing the value of sports data. In our first session, our Chief Scientist, Patrick Lucey, PHD explored the past, present and future of sports analytics.
Hoang Le, Peter Carr, Yisong Yue and Patrick Lucey’s 2017 entry titled “Data-Driven Ghosting Using Deep Imitation Learning” utilised player and ball tracking to analyse player decision making, specifically in defensive situations.
Our work showcased an automatic “data-driven ghosting” method using advanced machine learning methodologies called “deep imitation learning”, applied to a season’s worth of tracking data from a recent professional league in soccer. Our ghosting method, which avoids substantial manual human annotation, results in a data-driven system that allowed us to answer the question “how should this player or team have played in a given game situation compare to the league average?”
Our team members Panna Felson, Sujoy Ganguly, and Patrick Lucey’s presented our paper “Where Will They Go? Predicting Fine-Grained Adversarial Multi-Agent Motion using Conditional Variational Autoencoders” at the 2018 European Conference on Computer Vision.
In our paper, we presented a technique using conditional variational auto-encoder which learns a model that “personalises” prediction to individual agent behaviour within a group representation. Given the volume of data available and its adversarial nature, we focused on the sport of basketball to show that our approach efficiently predicts context-specific agent motions. We found that our model generates results that are three times as accurate as previous state-of-the-art approaches (5.74 ft vs. 17.95 ft).
Each year Stats Perform’s AI team attends MIT’s Sloan Sports Conference to showcase our latest innovations in sports technology. Our 2017 entry from Panna Felson and Patrick Lucey titled ““BodyShots”: Analysing Shooting Styles in the NBA using 3D Body-Pose Information” developed a novel attribute-based representation of basketball player’s body pose during a three point shot.
Using Pearson’s Chi-squared test to quantify differences in attribute distributions for made and missed shots, we observed statistically significant differences in distributions of attributes describing the style of movement, e.g., walk, run, or hop, in various phases of the shot.
The paper includes a case study on Stephen Curry, where we observed that Curry moves much more than the average player in all phases of his shot and he takes a higher proportion of off-balance shots compared to the average player.
Sketching plays is a universal way for coaches to communicate what they want their players to do. What if a coach didn’t have to rely solely on intuition, but could instead foresee how the defending team is likely to respond to the intended play? For the 2018 MIT Sloan Sports Conference Thomas Seidl, Aditya Cherukumudi, Andrew Hartnett, Peter Carr and Patrick Lucey did just that.
In our work, we considered play sketching from a data-driven perspective. We combined a powerful analytics framework built on deep-imitation learning with an intelligent and highly intuitive user interface. Users freehand sketch plays or modify existing tracking data. Our software then inferred the equivalent animation and synthesised realistic “ghost” defenders. Users tested their plays against different teams and game contexts, and fine-tune sketches to maximise the expected points in a given situation.
MIT SLOAN SPORTS CONFERENCE
During the 2016-17 Premier League season, 16% of all goals scored came from corners and free kicks. However some teams proved more adept than others at generating opportunities from these situations.
Stats Perform’s 2018 Research Paper, titled ‘Mythbusting Set-Pieces in Soccer’ was co-authored by Paul Power, Jennifer Hobbs, Hector Ruiz, Xinyu Wei and Patrick Lucey and introduced a novel, imaged-based representation approach for analysing set-piece situations.
Using a combination of tracking and event data, the paper presents a framework for detecting a team’s defensive set-piece structure, with the objective of exploiting inefficiencies and maximising goalscoring opportunities.
Stats Perform’s 2020 Research Paper, titled “You Cannot Do That Ben Stokes: Dynamically Predicting Shot Type in Cricket Using a Personalised Deep Neural Network”, was shortlisted as a finalist in the research track at the MIT Sloan Sports Analytics Conference.
Co-authored by Will Gürpınar-Morgan, Daniel Dinsdale, Joe Gallagher, Aditya Cherukumudi and Patrick Lucey, the paper introduced a new model for dynamically predicting a batsman’s shot type in One Day International cricket, using ball-by-ball event data.
The model takes into account various factors, including the game state, the bowler’s delivery trajectory and various personalised metrics for the batsman to generate probabilities of shot type from each ball and their-end location, highlighting the most likely outcome.
MIT SLOAN SPORTS CONFERENCE
In elite soccer, the aim of a team’s defence is to make the opposition’s play predictable. So if the art of good defending is to make play predictable, then it should be measurable.
Stats Perform’s 2021 Research Paper, titled ‘Making Offensive Play Predictable – Using a Graph Convolutional Network to Understand Defensive Performance in Soccer’ was co-authored by Paul Power, Michael Stöckl, Thomas Seidl and Daniel Marley and shortlisted as a finalist in the research track at the MIT Sloan Sports Analytics Conference.
Using tracking data, the paper outlines how it is possible to accurately model defensive behaviour and measure its effect on an opponent’s attacking behaviour.
Late in the 2018 Champions League Final, a key substitution resulted in an 83rd-minute goal to give Real Madrid the victory over Liverpool. The match result prompted several world-record-breaking goalkeeper transfers across multiple European leagues.
The high-profile transfers prompted the question, “how can you compare the performance of different goalkeepers across teams and leagues, when they don’t have head-to-head matchups or common opponents?”
Our AI team utilised spatial and body-pose data to bring deep, objective insights to the conversation and determine the best fit for clubs.
Learn more about Stats Perform’s AI-powered solutions.
