Outplaying opponents—a differential perspective on passes using position data

Shortcomings of applying data science to improve professional football performance: Takeaways from a pilot intervention study

Positional tracking data allows football practitioners to derive features that describe patterns of player behavior and quantify performance. Existing research using tracking data has mostly focused on what occurred on the pitch, such as the determinants of effective passing. There have yet to be studies attempting to use findings from data science to improve performance. Therefore, 24 professional players (mean age = 21.6 years, SD = 5.7) were divided into a control team and an intervention team which competed against each other in a pre-test match. Metrics were gathered via notational analysis (number of passes, penalty box entries, shots on goal), and positional tracking data including pass length, pass velocity, defensive disruption (D-Def), and the number of outplayed opponents (NOO). D-Def and NOO were used to extract video clips from the pre-test that were shown to the intervention team as a teaching tool for 2 weeks prior to the post-test match. The results in the post-test showed no significant improvements from the pre-test between the Intervention Team and the Control Team for D-Def (F = 1.100, p = 0.308, η² = 0.058) or NOO (F = 0.347, p = 0.563, η² = 0.019). However, the Intervention Team made greater numerical increases for number of passes, penalty box entries, and shots on goal in the post-test match. Despite a positive tendency from the intervention, results indicate the transfer of knowledge from data science to performance was lacking. Future studies should aim to include coaches' input and use the metrics to design training exercises that encourage the desired behavior.

Reliability of spatial-temporal metrics used to assess collective behaviours in football: An in-silico experiment

The purpose of this study was to investigate the reliability of spatial-temporal measurements applied within collective behaviour research in football. In-silico experiments were conducted introducing positional errors (0.5, 2 and 4 m) representative of commercial tracking systems to match data from the 2020 European Championship qualifiers. Ratios of the natural variance ("signal") of spatial-temporal metrics obtained throughout sections of each game relative to the variance created by positional errors ("noise") were taken to calculate reliability. The effects of error magnitude and time of analysis (1, 5 and 15 mins; length of attack: <10, 10-20, >20 s) were assessed and compared using Cohen's f² effect size. Error magnitude was found to exert greater influence on reliability (f² = 0.15 to 0.81) compared with both standard time of analysis (f² = 0.03 to 0.08) and length of attacks (f² = 0.15 to 0.32). the results demonstrate that technologies generating positional errors of 0.5 m or less should be expected to produce spatial-temporal metrics with high reliability. However, technologies that generate errors of 2 m or greater may produce unreliable values, particularly when analyses are conducted over discrete events such as attacks, which although critical, are often short in duration.

Off-ball behavior in association football: A data-driven model to measure changes in individual defensive pressure

This study describes an approach to evaluate the off-ball behaviour of attacking players in association football. The aim was to implement a defensive pressure model to examine an offensive player's ability to create separation from a defender using 1411 high-intensity off-ball actions including 988 Deep Runs (DRs) DRs and 423 Change of Directions (CODs). Twenty-two official matches (14 competitive matches and 8 friendlies) of the German National Team were included in the research. To validate the effectiveness of the pressure model, each pass (n = 25,418) was evaluated for defensive pressure on the receiver at the moment of the pass and for the pass completion rate (R = -.34, p < .001). Next, after assessing the inter-rater reliability (Fleiss Kappa of 80 for DRs and 78 for CODs), three expert raters annotated all DRs and CODs that met the pre-set criteria. A time-series analysis of each DR and COD was calculated to the nearest 0.1 second, finding a slight increase in pressure from the start to the end of the off-ball actions as defenders re-established proximity to the attacker after separation was created. A linear mixed model using run type (DR or COD) as a fixed effect with the local maximum as a fixed effect on a continuous scale resulted in p < 0.001, d = 4.81, CI = 0.63 to 0.67 for the greatest decrease in pressure, p < 0.001, d = 0.143, CI = 9.18 to 10.61 for length of the longest decrease in pressure, and p < 0.001, d = 1.13, CI = 0.90 to 1.11 for the fastest rate of decrease in pressure. As these values pertain to the local maximum, situations with greater starting pressure on the attacker often led to greater subsequent decreases. Furthermore, there was a significant (p < .0001) difference between offensive and defensive positions and the number of off-ball actions. Results suggest the model can be applied to quantify and visualise the pressure exerted on non-ball-possessing players. This approach can be combined with other methods of match analysis, providing practitioners with new opportunities to measure tactical performance in football.

Expected passes

Passes are by far football’s (soccer) most frequent event, yet surprisingly little meaningful research has been devoted to quantify them. With the increase in availability of so-called positional data, describing the positioning of players and ball at every moment of the game, our work aims to determine the difficulty of every pass by calculating its success probability based on its surrounding circumstances. As most experts will agree, not all passes are of equal difficulty, however, most traditional metrics count them as such. With our work we can quantify how well players can execute passes, assess their risk profile, and even compute completion probabilities for hypothetical passes by combining physical and machine learning models. Our model uses the first 0.4 seconds of a ball trajectory and the movement vectors of all players to predict the intended target of a pass with an accuracy of $$93.0\%$$ 93.0 % for successful and $$72.0\%$$ 72.0 % for unsuccessful passes much higher than any previously published work. Our extreme gradient boosting model can then quantify the likelihood of a successful pass completion towards the identified target with an area under the curve (AUC) of $$93.4\%$$ 93.4 % . Finally, we discuss several potential applications, like player scouting or evaluating pass decisions.

The origins of goals in the German Bundesliga

We propose to analyse the origin of goals in professional football (soccer) in a purely data-driven approach. Based on positional and event data of 3,457 goals from two seasons German Bundesliga and 2nd Bundesliga (2018/20,219 and 2019/2020), we devise a rich set of 37 features that can be extracted automatically and propose a hierarchical clustering approach to identify group structures. The results consist of 50 interpretable clusters revealing insights into scoring patterns. The hierarchical clustering found 8 alone standing clusters (penalties, direct free kicks, kick and rush, one-two's, assisted by header, assisted by throw-in) and nine categories (e.g., corners) combining more granular patterns (e.g., five subcategories of corner-goals). We provide a thorough discussion of the clustering and show its relevance for practical applications in opponent analysis, player scouting and for long-term investigations. All stages of this work have been supported by professional analysts from clubs and federation.

Data-driven detection of counterpressing in professional football

Detecting counterpressing is an important task for any professional match-analyst in football (soccer), but is being done exclusively manually by observing video footage. The purpose of this paper is not only to automatically identify this strategy, but also to derive metrics that support coaches with the analysis of transition situations. Additionally, we want to infer objective influence factors for its success and assess the validity of peer-created rules of thumb established in by practitioners. Based on a combination of positional and event data we detect counterpressing situations as a supervised machine learning task. Together, with professional match-analysis experts we discussed and consolidated a consistent definition, extracted 134 features and manually labeled more than 20, 000 defensive transition situations from 97 professional football matches. The extreme gradient boosting model—with an area under the curve of $$87.4\%$$ 87.4 % on the labeled test data—enabled us to judge how quickly teams can win the ball back with counterpressing strategies, how many shots they create or allow immediately afterwards and to determine what the most important success drivers are. We applied this automatic detection on all matches from six full seasons of the German Bundesliga and quantified the defensive and offensive consequences when applying counterpressing for each team. Automating the task saves analysts a tremendous amount of time, standardizes the otherwise subjective task, and allows to identify trends within larger data-sets. We present an effective way of how the detection and the lessons learned from this investigation are integrated effectively into common match-analysis processes.

Attacking Key Performance Indicators in Soccer: Current Practice and Perceptions from the Elite to Youth Academy Level

Key Performance Indicators (KPIs) are used to evaluate the offensive success of a soccer team (e.g. penalty box entries) or player (e.g. pass completion rate). However, knowledge transfer from research to applied practice is understudied. The current study queried practitioners (n = 145, mean ± SD age: 36 ± 9 years) from 42 countries across different roles and levels of competition (National Team Federation to Youth Academy levels) on various forms of data collection, including an explicit assessment of twelve attacking KPIs. 64.3% of practitioners use data tools and applications weekly (predominately) to gather KPIs during matches. 83% of practitioners use event data compared to only 52% of practitioners using positional data, with a preference for shooting related KPIs. Differences in the use and value of metrics derived from positional tracking data (including Ball Possession Metrics) were evident between job role and level of competition. These findings demonstrate that practitioners implement KPIs and gather tactical information in a variety of ways with a preference for simpler metrics related to shots. The low perceived value of newer KPIs afforded by positional data could be explained by low buy-in, a lack of education across practitioners, or insufficient translation of findings by experts towards practice.