Sports Analytics Weekly by kubeia.io - 17/2024

The Toronto Raptors’ Jontay Porter was just banned for life for violating betting rules. There will be much more of this to come

Next week we might witness NFL Draft history. Current projections have zero running backs or linebackers taken in the 1st round. This has never happened before in the 89-year history of the draft. There’s a lot going on here, but as the great Ludacris says, have we seen the last of a dying breed?

Here we tackle (pun intended) the mathematics behind timing the perfect tackle! Turns out with some algebra, trigonometry, and optimization we can know exactly what angle to run at in order to intercept a wide receiver, and coaches actually use some of these tactics when training their players!

In the dynamic landscape of modern basketball, analytics has emerged as a cornerstone, fundamentally altering the way the game is played, analyzed, and strategized. This panel will delve into the emergence of basketball analytics from its initiation to its current essential role in today's basketball paradigm. The discussion seamlessly intertwines the historical origins with the contemporary significance of basketball analytics, showcasing its transformative journey within the sport, from its humble origins to its current status as a game-changer in the NBA and professional basketball. Navigating pivotal milestones, influential figures, and driving forces, the panel examines the profound impact analytics has on various facets of the sport, from front office strategies, coaching methodologies, player development, and beyond. This session promises to captivate enthusiasts, industry professionals, and basketball aficionados alike, offering valuable insights into the significance of analytics in shaping the past, present, and future of basketball strategy and performance.

We develop a novel type of basketball shot chart, a true shot chart, that uses a generalized additive model (GAM) to estimate total shot proficiency continuously in the half-court as a continuous, 3-D surface (https://sportdataviz.syr.edu/TrueShotChart/). Unlike conventional shot charts, which do not consider free throw scoring pursuant to a shot from a given location, true shot charts incorporate total points, from the field and free throw line, pursuant to each shot in an NBA game (from 2016-2022 in the study) toward improved explanatory power of offensive efficiency variation across NBA team-seasons. Whereas conventional shot charts show a league-wide three-point premium over the period of the data, true shot charts show a deepening dispremium since 2018, as the free throw rate for three-point attempts is substantially less than that for two-point attempts. Lastly, we develop a novel shot chart summary measure, shot selection efficiency, as the Pearson correlation between expected proportional volume and expected true points, from the field and free throw line, across the half court space; polynomial regression and XGBoost modeling suggest shot selection efficiency is not only win productive, but a “Moneyball” or partly supra-payroll source of wins.  ‍

In spite of the importance of the pass rush in professional football, pass rushing statistics only include the final outcomes of a play, e.g., sack and pass-made. They do not capture the dynamics of the pass rush or fine-grained insights throughout a play on how much pressure a rusher generates during the rush. In this paper, we propose a unified framework that enables the estimation of defensive pressure scores throughout a play with high accuracy and performance for rusher and blocker identification, rusher-blocker match-up and pressure score estimation and show the real-world applications of our framework including enriched analytics.

Impacting the Game: The Advancement of Sports Analytics with Industry Pioneers, presented by ESPNESPN Analytics will host a conversation with two leaders in the advanced metrics space, hitting on the prominence of analytics in the recently completed NFL playoffs, the impact on football and basketball in-game decision making, the media’s role in presenting analytics, and much more.

Although the sports industry pours millions of dollars into understanding talent, we do not know: how to measure individuals’ attitudes towards competition, when these attitudes are formed, how they vary both within individuals over time and across individuals, and, more fundamentally, how important competitiveness is for sporting success. We measure competitiveness and answer these questions by leveraging a rich, dynamic panel dataset on hundreds of top young prospects from a renowned professional soccer academy during the decade leading up to professionalism. The ideas and methods are applicable to all other sports.

This paper presents a novel approach to positioning baseball fielders to maximize expected outs or minimize expected runs allowed against an opposing hitter. We find evidence that our positioning approach is an improvement over MLB average positioning in terms of both hits and runs allowed. We then extend our approach to adaptable hitters who adjust their batted ball strategy in response to the defense’s positioning strategy by modeling the interaction as a zero-sum game and solving for an equilibrium pair of strategies. We demonstrate two examples where the game theory model is appropriate: against shift-beating hitters who pull the ball less frequently when the defense shifts against them and against pull-heavy left-handed hitters who threaten to bunt against an extreme shift.

Player injuries in soccer significantly impact team performance, club financial stability and player welfare, with the ‘Big Five’ European soccer leagues experiencing a staggering £513 million in injury-related costs during the 2021/22 season. In this paper, we present a novel forward-looking team selection model, framed as a Markov decision process and optimised with Monte Carlo tree search, that balances team performance with the risk of long-term player unavailability due to injury. We demonstrate that real-world teams could reduce the incidence of player injury by ~13% and wages inefficiently spent on injured players by ~11% using our data-driven team selection model.

We introduce a new mathematical system for predicting outcomes of NBA draft prospects based on the outcomes of other previously drafted players. This approach, which is completely general and applicable to any sport, forms predictions as relevance-weighted averages of prior outcomes using a precise and theoretically justified assessment of relevance derived from principles of information theory. Crucially, a measure called “fit” indicates in advance the unique reliability of each individual prediction and dynamically focuses each prediction on the combinations of predictive variables and previous players that are most informative for the prediction task. Relevance-based prediction addresses complexities that are beyond the capacity of conventional prediction models, but in a way that is more transparent, more flexible, and more theoretically justified than widely used machine learning algorithms.

This paper presents a novel machine learning method to measure the pressure in soccer games. We first propose a technique to quantify the pressure on individual player with 3D body motion parameters considered. Compared to the vanilla approach, our 3D quantification method provides more accurate results. Based on the individual assessment of each player, we propose player pressure map (PPM) to represent a given game scene, which lowers the dimension of raw data and still contains rich contextual information. We then train a possession outcome prediction model on PPMs and use the predicted probability to quantify the pressure on the whole team. This quantification enables contextualized performance analysis with team pressure taken into consideration. Overall, our model provides coaches and analysts with an efficient tool to quantify the pressure at both individual and team levels, which can help them evaluate the team performance more precisely.

In hockey, faceoffs have long been acknowledged as important drivers of puck possession, but their actual impact on scoring outcomes remains inadequately measured. It is acceptedly evident that a center winning 54% of their faceoffs outperforms one with a 51% success rate, but the tangible extent of this advantage in terms of goals, wins, and losses remains underexplored. This research fills the void by continuing the effort to translate faceoff results to scoring outcomes, measuring faceoff performance in goals, wins, and losses in a novel manner. We explore evidence that faceoffs are an undervalued championship-caliber market inefficiency and offer models enabling General Managers to see role-specific projections of how different personnel and usage could maximize offense, defense, and championship chances.‍

Player tracking data can enhance the quantification of player abilities and our understanding of team composition and broader team strategies. In this work, we demonstrate how density-functional fluctuation theory (DFFT), an extension of a Nobel Prize-winning physics approach, can process basketball tracking data by treating players as interacting densities. By training the interactions on different play outcomes, we can evaluate play-outcome likelihoods based on player positions, determine which players are in strong or weak positions, and understand which players consistently instigate strong responses from the opposing team (i.e., ‘player gravity’). We find that our approach not only identifies the overall strengths of a player, but also identifies subtleties such as those who are left-handed (e.g., D. Russell) or who instigate changes non-locally through frequent passes (e.g., N. Jokic). 

This panel will deep dive into the multi-faceted world of football analytics with seasoned industry experts from the Washington Commanders, Pro Football Focus, SumerSports, and StatsBomb. The panelists will discuss the current breadth of analytics use cases in football, ranging from the crucial role data plays in player evaluation and roster construction to the rapidly increasing utilization of data to inform game strategy and decision-making. The panelists will highlight the tangible impacts of data-driven methodologies in shaping the modern football landscape and explore potential analytics advancements that can further enhance the influence of data on success in the NFL.

Three is worth more than two, walks count the same as hits, first downs are better than punts. While data analysis has transformed the way we play all of the major American sports, soccer still awaits its own kind of disruption. It’s not for a lack of trying, though. Major European clubs have begun to invest — and occasionally empower — in data-driven decision-making. But due to cultural differences, institutional inertia, and a sport that’s simply really hard to measure, advanced analytics haven't had the same kind of transformative effect at the highest level of the sport that we've seen in the NBA, MLB, or NFL. Combining on-the-ground coaching expertise with high-level quantitative thinking, this panel will survey the state of soccer’s analytical revolution and attempt to determine just how far it might go.

There's a changing of the guard underway in the NHL, with legendary superstars entering the twilight of their careers and a new crop of generational talents on the way up. As this new era of the NHL gets underway, how are front office executives thinking about roster construction? What's the balance between analytics and eye test? How do you construct a roster that can handle the grind of the regular season while making sure you have a team that's built to succeed in the playoffs? From the player’s perspective, how is this next generation of NHL athletes thinking about analytics to improve their game, scout opposing teams and maximize their recovery? Come find out all this and more from 13 year NHL veteran P.K. Subban, NHL front office executives Brett Peterson and Tyler Delllow and Stathletes Co-Founder and CEO Meghan Chayka.

Research Paper Title: The Strain of Success: A Predictive Model for Injury Risk Mitigation and Team Success in SoccerAbstract: Player injuries in soccer significantly impact team performance, club financial stability and player welfare, with the ‘Big Five’ European soccer leagues experiencing a staggering £513 million in injury-related costs during the 2021/22 season. In this paper, we present a novel forward-looking team selection model, framed as a Markov decision process and optimised with Monte Carlo tree search, that balances team performance with the risk of long-term player unavailability due to injury. We demonstrate that real-world teams could reduce the incidence of player injury by ~13% and wages inefficiently spent on injured players by ~11% using our data-driven team selection model.

While the tactical behavior of soccer players differs between specific phases of play (offense, defense, offensive transition, defensive transition), little is known about successful behavior of players during defensive transition (switching behavior from offense to defense). Therefore, this study aims to analyze the group tactic of rest defense (despite in ball possession, certain players safeguard quick counterattacks in case of ball loss) in defensive transition. A mixed-methods approach was used, involving both qualitative and quantitative analysis. Semi-structured expert interviews with seven professional soccer coaches were conducted to define rest defense. In the quantitative analysis, several KPIs were calculated, based on tracking and event data of 153 games of the 2020/21 German Bundesliga season, to predict the success of rest defense situations in a machine learning approach. The qualitative interviews indicated that rest defense can be defined as the positioning of the deepest defenders during ball possession to prevent an opposing counterattack after a ball loss. For instance, the rest defending players created a numerical superiority of 1.69 ± 1.00 and allowed a space control of the attacking team of 11.51 ± 9.82 [%] in the area of rest defense. The final machine learning model showed satisfactory prediction performance of the success of rest defense (Accuracy: 0.97, Precision: 0.73, f1-Score: 0.64, AUC: 0.60). Analysis of the individual KPIs revealed insights into successful behavior of players in rest defense, including controlling deep spaces and dangerous counterattackers. The study concludes regaining possession as fast as possible after a ball loss is the most important success factor in defensive transition.

The interest in tactical analysis in soccer has increased in the latest years, especially with the growing availability of player tracking data. With it, the defending team's compact organization, which is considered by practitioners to be an important factor in defense, was repeatedly examined. However, the connection between this defensive principle of play and the defending success remains unclear. Therefore, this study aims to investigate the relation of the principle of playing defensive compact organization to the success of the defense. Based on tracking and event data of 153 games of the German Bundesliga (season 2020/21), the compactness (surface area, spread of the team, and of defending subgroups) and the organization (distances between formation lines) of the defending team was compared between successful and unsuccessful defensive plays. There were almost no differences in the compactness of the whole team, and the organizational measures between successful and unsuccessful defensive plays. The defending subgroup of five defenders closest to the ball showed a higher compactness (smaller surface area and smaller spread) in successful defensive plays compared to unsuccessful ones (−0.08 ≤ d ≤ −0.16). Our results indicate that the compactness of players in areas close to the ball seems crucial for defensive success. However, the compact organization of the entire team does not seem important to regain the ball in defense.

The tactical formation has been shown to influence the match performance of professional soccer players. This study aimed to examine the effects of in-game changes in tactical formation on match performance and to analyze coach-specific differences. We investigated three consecutive seasons of an elite team in the German Bundesliga which were managed by three different coaches, respectively. For every season, the formation changes that occurred during games were recorded. The match performance was measured on a team level using the variables “goals,” “chances,” and “scoring zone” entries (≙successful attacking sequence) for the own/opposing team. Non-parametric tests were used to compare the 10 min before with the 10 min after the formation change, as well as games with and without formation change. In the 10 min after the formation change, the team achieved more goals/chances/scoring zone entries than in the 10 min before the formation change (mean ES = 0.52). Similarly, the team conceded fewer opposing goals/chances/scoring zone entries in the 10 min after the formation change (mean ES = 0.35). Furthermore, the results indicate that the success of the respective formation change was dependent on the responsible coach. Depending on the season, the extent of the impacts varied (season 1: mean ES = 0.71; season 2: mean ES = 0.26; and season 3: mean ES = 0.22). Over all three seasons, the formation changes had a positive effect on the match performance of the analyzed team, highlighting their importance in professional soccer. Depending on the season, formation changes had varying impacts on the performance, indicating coach-specific differences. Therefore, the quality of the formation changes of the different coaches varied. The provided information can support coaches in understanding the effects of their in-game decisions.

Deep learning web application for football analysis with Streamlit.

Over the last 25 years, soccer tracking data has provided a deeper understanding of the ways that players and teams play the game. Although traditional tracking systems require in-venue installation, there is a current push to track players remotely from broadcast footage. However, tracking data obtained from broadcast footage is inherently incomplete due to players being out of the broadcast camera’s field of vision. We address this issue in this paper, leveraging generative AI to predict highly accurate locations of the players for the large portions of games where they cannot be visually perceived.

American football has in recent years made drastic shifts towards the quantitative. The proliferation of charting and tracking data has given us the ability to better evaluate and value players, but player development has been left wanting. In this paper we use NFL's Next Gen Stats data to build tools for offensive linemen in pass protection, which will help teams more efficiently watch film, monitor performance and fitness, build rosters, and game plan, with player development as the central focus. 

Algorithmic Trading and Quantitative Strategy Lecture 7 is posted. BACKTESTING.

Portfolio optimization chiefly focuses on risk and return prediction, yet implementation costs also play a critical role. Predicting trading costs is challenging, however, since costs depend endogenously on trade size and trader identity, thus impeding a generic solution. We focus on a key, yet general, component of trading costs that abstracts from these challenges---trading volume. Trading volume is highly predictable, especially with machine learning. We model the economic benefits of predicting volume through a portfolio framework that trades off portfolio tracking error versus net-of-cost performance---in essence translating volume prediction into net-of-cost portfolio alpha. We find the benefits of predicting volume to be substantial, and potentially as large as those from return prediction.

Common Machine Learning applications in sports analytics relate to player injury prediction and prevention, potential skill or market value evaluation, as well as team or player performance prediction. This paper focuses on football. Its scope is long–term team and player performance prediction. A reliable prediction of the final league table for certain leagues is presented, using past data and advanced statistics. Other predictions for team performance included refer to whether a team is going to have a better season than the last one. Furthermore, we approach detection and recording of personal skills and statistical categories that separate an excellent from an average central defender. Experimental results range between encouraging to remarkable, especially given that predictions were based on data available at the beginning of the season.

On scholarship in the statistics community, trend filtering, conformal prediction, and epidemic forecasting.

Imbalance correction led to models with strong miscalibration without better ability to distinguish between patients with and without the outcome event. The inaccurate probability estimates reduce the clinical utility of the model, because decisions about treatment are ill-informed.

This document contains derivations and algorithms for implementing Hidden Markov Models. The content presented here is a collection of my notes and personal insights from two seminal papers on HMMs by Rabiner in 1989 [2] and Ghahramani in 2001 [1], and also from Kevin Murphy’s book [3]. This is an excerpt from my project report for the MIT 6.867 Machine Learning class taught in Fall 2014.