Players’ decisions within ball screens in elite German men’s basketball: observation of offensive–defensive interactions using a process-orientated state-event model

HoopTransformer: Advancing NBA Offensive Play Recognition with Self-Supervised Learning from Player Trajectories

BACKGROUND AND OBJECTIVE

Understanding and recognizing basketball offensive set plays, which involve intricate interactions between players, have always been regarded as challenging tasks for untrained humans, not to mention machines. In this study, our objective is to propose an artificial intelligence model that can automatically recognize offensive plays using a novel self-supervised learning approach.

METHODS

The dataset was collected by SportVU from 632 games during the 2015-2016 season of the National Basketball Association (NBA), with a total of 90,524 possessions. A multi-agent motion prediction pretraining model was built on the basis of axial-attention transformer and trained with different masking strategies: motion prediction (MP), motion reconstruction (MR), and MP + MR joint strategy. A downstream play-level classification task and similarity search were used to evaluate the models' performance.

RESULTS

The results showed that the MP + MR joint masking strategy maximized the ability of the model compared with individual masking strategies. For the classification task, the joint strategy achieved a top-1 accuracy of 81.5% and top-3 accuracy of 97.5%. In the similarity search evaluation, the joint strategy attained a top-5 accuracy of 76% and top-10 accuracy of 59%. Additionally, with the same MP + MR joint masking strategy, our HoopTransformer model outperformed the two baseline models in the classification task and similarity search.

CONCLUSION

This study presents a self-supervised learning model and demonstrates the effectiveness and potential of the model in accurately comprehending and capturing player movements and complex interactions during offensive plays.

Evaluating field-goal shooting effectiveness in wheelchair basketball players across a competitive season: a preliminary study

Background

Information about non-elite wheelchair basketball (WB) players across national competitive seasons are still missing. This study aimed at identifying which situational-related variables were associated with shooting effectiveness in non-elite WB players.

Methods

All the matches played by one WB team across one national competitive season were video-recorded and analysed; 333 shooting attempts from high-point players and several situational-related variables were considered.

Results

Pearson's Chi-square test showed that increased shooting effectiveness under the following conditions: playing on home ground, during won matches, while taking shots with the wheelchair in motion, and when no opposing player raised their arm in defence. Results of the multivariable logistic regression analysis showed a statistically significant influence of match location (p-value = 0.001), shot-clock remaining (p-value = 0.015) and modality of press (p-value < 0.001). The highest attack effectiveness was achieved when teams played at home (odds ratio [OR] = 2.49), while the shooting effectiveness decreased when the shot occurred during the last seconds of the action (OR = 0.36), or the opponents defended with the arm raised (OR = 0.19). These results suggest that coaches should include exercises aimed at shooting under conditions of increased pressure in their programmes in order to create specific situations during the training sessions to prepare their high-point athletes for shots under specific match constraints.

Ball Screens in the Men's 2019 Basketball World Cup

BACKGROUND

The objective of this research was to analyse the effectiveness and search for successful patterns in ball screens in the men's 2019 Basketball World Cup.

METHODS

The sample consisted of 515 ball screens obtained in critical moments. LINCE software was used as a registration instrument by means of an observational instrument designed ad hoc. A descriptive analysis and chi-square tests (χ2) were performed with SPSS 25.0 and a T-patterns analysis with Theme 5 software.

RESULTS

The results indicate that the criteria that have the most influence on this type of action are the result of the team executing the screen (winning, losing or tying), the type of offense and the defence used on the ball screen. The most representative patterns of success tend to take place between 9-16 s of possession, with the screeners being inside players and the screened players being outside players, and it is performed in the upper areas of the court against an individual type of defence and ending with the screened player advancing towards the basket or passing to an open teammate.

CONCLUSIONS

The data obtained will enable the coaching staff to train ball screens in accordance with specific game situations.

The Pick-and-Roll in Basketball From Deep Interviews of Elite Coaches: A Mixed Method Approach From Polar Coordinate Analysis

Pick-and-roll is the most widespread cooperative action among high-level basketball teams and the most applied strategy by coaches to gain an advantage over the rival team. During pick-and-roll, opposing teams perform antagonistic actions based on goals that are expressed in offensive and defensive tactics. The aim of this study is to examine the approaches of high-level coaches on the offensive and defensive dynamics emerging in matches of a basketball elite team during an entire season of the Spanish Asociación de Clubes de Baloncesto (ACB) league. To this end, we used a mixed-methods approach based on systematic observation of verbatim transcripts of interviews conducted with six high-level coaches about the pick-and-roll dynamics that emerged in matches of the Unicaja Málaga team during an entire season of the ACB league. The observational design was nomothetic, punctual, and multidimensional. The choice of this methodology is justified since we developed an ad hoc indirect observation tool to evaluate the coaches’ perspective on this dynamic. Once the intra-observer reliability of the instrument was confirmed, we performed a polar coordinate analysis to identify the significant relationships between the coaches’ evaluations and the offensive and defensive pick-and-roll elements that supported such verbal behaviors. The results highlight the presence of various offensive and defensive aspects of pick-and-roll (n = 2224) emerging in the Unicaja team that were significantly associated with positive and negative evaluations of the coaches. The interview confirms that coach 1 and his staff were less confident in options that pick-and-roll offer, which is also reflected in the record of screens made and simulated, than coach 3. This study shows that the application of mixed methods, by analysis of the polar coordinate of the coding carried out on responses of a systematized interview, has proven to be an effective strategy in obtaining relevant information on the expert knowledge of the elite coaches on the influence of pick-and-roll on tactical actions in basketball.

Effect of Ball Screen and One-on-One on the Level of Opposition and Effectiveness of Shots in the ACB

In light of the importance of shots on the final result in basketball matches, this research aims to reveal the predictive ability of variables associated with different game actions (ball screen, one-on-one near the basket and one-on-one far away from the basket), shot zone (near to or far from the basket) and players' position (outside or inside) of the success of shots and of the type of shot, defended or not defended, in three seasons (2014 to 2016) of the Copa del Rey in the ACB (Asociación de Clubes de Baloncesto) League in Spain. Observational methodology was used, and an ad hoc instrument was created using Lince software. The study was conducted on a total of 7695 game actions, of which an analysis of 877 shots was made using CHAID (Chi-squared Automatic Interaction Detection) classification trees. The results obtained in this research have led us to the following conclusions: (a) the most important variable in the success of shots is to find "open shots", regardless of the game action, (b) one-on-one actions outside and inside the zone have a clear connection to the finishes of the plays and end up in defended shots, (c) shots by ball screen lead to more open shots than in one-on-one and (d) play with passes may facilitate to find "open shots". These results can be of interest for coaches to prepare their teams in the best possible way, but they must be interpreted with caution as the sample was very specific (Copa del Rey in the ACB League) and, therefore, it is necessary to continue investigating this topic in other competition contexts.

Predicting field-goal success according to offensive, defensive and contextual variables in elite men's wheelchair basketball

The purposes of this study were to (i) develop a field-goal shooting performance analysis template and (ii) explore the impact of each identified variable upon the likely outcome of a field-goal attempt using binary logistic regression modelling in elite men’s wheelchair basketball. First, a field-goal shooting performance analysis template was developed that included 71 Action Variables (AV) grouped within 22 Categorical Predictor Variables (CPV) representing offensive, defensive and game context variables. Second, footage of all 5,105 field-goal attempts from 12 teams during the men’s 2016 Rio De Janeiro Paralympic Games wheelchair basketball competition were analysed using the template. Pearson’s chi-square analyses found that 18 of the CPV were significantly associated with field-goal attempt outcome (p < 0.05), with seven of them reaching moderate association (Cramer’s V: 0.1–0.3). Third, using 70% of the dataset (3,574 field-goal attempts), binary logistic regression analyses identified that five offensive variables (classification category of the player, the action leading up to the field-goal attempt, the time left on the clock, the location of the shot, and the movement of the player), two defensive variables (the pressure being exerted by the defence, and the number of defenders within a 1-meter radius) and 1 context variable (the finishing position of the team in the competition) affected the probability of a successful field-goal attempt. The quality of the developed model was determined acceptable (greater than 65%), producing an area under the curve value of 68.5% when the model was run against the remaining 30% of the dataset (1,531 field-goal attempts). The development of the model from such a large sample of objective data is unique. As such it offers robust empirical evidence to enable coaches, performance analysts and players to move beyond anecdote, in order to appreciate the potential effect of various and varying offensive, defensive and contextual variables on field-goal success.

Analyse der Interaktionen im direkten Block als vorbereitende und abschließende Angriffshandlung in zufällig ausgewählten Hauptrundenspielen der easyCredit Basketball Bundesliga 2017/18

Der direkte Block (engl. „ball screen“, meist als Pick & Roll mit Abrollen des Blockstellers zum Korb ausgeführt) besitzt seit vielen Jahren einen hohen Stellenwert im Leistungsbasketball und wird in Positions- und Schnellangriffen gegen Mann-Mann- sowie Ball-Raum-Verteidigungsvarianten eingesetzt. Die vorliegende Studie analysiert die Häufigkeiten und Erfolgsquoten direkter Blocks zur Vorbereitung finaler Abschlusshandlungen und als direkte Abschlusshandlung anhand einer Zufallsstichprobe der Hauptrundenspiele der easyCredit Basketball Bundesliga 2017/18. Insgesamt 2555 Angriffssequenzen wurden mit einem an Remmert angelehnten Kategoriensystem mit Hilfe der Spielanalysesoftware utilius vs® detailliert beschrieben und quantitativ ausgewertet. Mit 19,9 % der Abschlusshandlungen und 26,4 % der vorbereitenden Abschlusshandlungen ist der quantitative Anteil der Blocks im Vergleich zu aktuellen internationalen Studien (siehe „Problemstellung“) niedrig. Die Erfolgsquote liegt jedoch mit durchschnittlich 1,22 Punkten pro Angriffsversuch (PPA) höher als die 1,16 PPA der 1‑gegen-1-Abschlüsse. 86,7 % der direkten Blocks zum Angriffsabschluss werden als aktive Blocks gestellt und sind mit 1,25 PPA überdurchschnittlich erfolgreich. Das Über-den-Block-Gehen und Switch sind die häufigsten Verteidigungsvarianten gegen direkte Blocks, erstere ist mit nur 1,08 zugelassenen PPA deutlich erfolgreicher. Die häufigste offensive Lösung im direkten Block ist der Wurf des Ballbesitzers, der mit 1,13 PPA jedoch nur mäßig lukrativ ist. Durchbruch und Abrollen sind mit 1,28 und 1,31 PPA effektiver. Gewinnerteams erzielen durchschnittlich 1,33 PPA aus direkten Blocks zum Angriffsabschluss, Verliererteams nur 1,10 PPA. Sie sind damit deutlich effektiver und verteidigen erfolgreicher.