ECSS Paris 2023: OP-AP08
INTRODUCTION: Indoor hockey is a small format version of field hockey, played with five outfield players and one goalkeeper per team. Matches consist of four 10-minute quarters, in a recommended 44-m (long) by 22-m (wide) area. Little is known of the match demands of this hockey format, despite obvious differences in team composition, match time and pitch size compared to outdoor field hockey. Local positioning system (LPS) technology, mapped to court parameters, allows the measurement of position in covered spaces, and hence allows the quantification of typical match demand metrics when performers are involved in indoor sport. METHODS: Following university ethical approval, 55 players, recruited across two tournaments, participated in the study (mean age: 24.2 (5.61) years; height: 1.7 (0.06) m; weight: 77.8 (9.72) kg). Match demands were assessed using a LPS (Catapult Sports S7 units 10 Hz, ClearSky system with 20 anchors, Catapult Sports, Melbourne, Australia). Heart rate was collected throughout matches (Polar H10, polar Electro, Kempele, Finland), and rating of perceived exertion was established immediately following the end of matches (modified Borg CR10 rate of perceived exertion (RPE) scale). Means and standard deviations were used to present the demands across 20 indoor matches (mean temperature: 18.9 (1.3) degrees Celsius; humidity: 43.7 (4.0) %) for the following metrics: total playing duration (min), total distance (m), high-speed running distance (m, greater than or equal to 15.5 km.h-1), accelerations (n, greater than or equal to 2 m.s-2), decelerations (n, less than or equal to - 2 m.s-2), average HR (bpm) and RPE (au). These metrics have been used to assess match demands in previous outdoor field hockey research (1). RESULTS: The demands of indoor hockey matches were as follows; total playing duration: 24.2 (5.6) min; total distance: 1984.1 (426.6) m; high-speed running distance: 136.0 (77.8) m; accelerations: 38.4 (15.8); decelerations: 38.5 (14.7); average HR: 170.7 (10.8) bpm; RPE: 5.5 (1.7). CONCLUSION: The data in this study builds the foundation for indoor hockey research using LPS technology, which has become commonplace in comparative indoor team sports. Access to this example data could give coaches and practitioners a better understanding of how to prepare their athletes for the demands of indoor hockey matches. [1] Lam 2021
Read CV Matt StonerECSS Paris 2023: OP-AP08
INTRODUCTION: Basketball is inherently dualistic often shaped by 1 vs 1 interactions of players. While basketball is on the forefront of sport analytics, match performance indicators don’t account for how and if the player is guarded by a defender. Automatically identifying defensive matchups is therefore a fundamental yet challenging task. Existing approaches are either simplistic, relying on the nearest-defender, or highly complex machine learning models, which achieves high accuracy but at considerable computational cost [1,2]. In this study, we aim to propose a feature-based optimization framework for defensive assignment and assess its performance against ground truth data. METHODS: SportVU tracking data from the 2015–2016 NBA regular season was used, providing ball and player positions at 25 Hz. A total of 1,228 possessions from 20 games are randomly selected, ending with shot attempts, standardized to a right-to-left attack, and downsampled to 1 Hz. We use motion-based features commonly applied in man-marking and ball pressures studies. After normalization, a cost function C_ij with initial weights from domain knowledge is built, where i denotes attacker and j defenders. Each frame’s cost matrix is converted to an assignment probability matrix P_ij. Then, feature weights are optimized by minimizing row and column entropy L using differential evolution. To evaluate model performance, an experienced sports analyst annotated defensive assignments for 20 random possessions by watching game footage. (1) C_ij = w1*I(i,ball-handler) + w2*d(i,basket) + w3*d(i,j) + w4*||v_i - v_j|| + w5*Delta_theta(i,j) (2) P_ij = exp(-C_ij) / sum_k exp(-C_ik) (3) L = sum_i H(P_i) + lambda sum_j H(P_j) H(p) = - sum_k p_k log(p_k) RESULTS: Our framework achieves an overall accuracy of 63.67% against expert-annotated ground truth. Among all features, distance between opponents has the largest absolute weight, making it the most influential factor in defensive assignment. Ball handler and velocity vector difference also play substantial roles. Attacker-basket distance contributes moderately, while angular difference has the smallest impact. CONCLUSION: Our study proposes an entropy-based soft optimization framework for defensive assignment, achieving 63.67% accuracy against expert labels. The study also shows that the location of the ball, players and the basket are 3 key elements for identifying defensive assignments, which places certain demands on the game dataset. By now accuracy remains below complex machine learning models, while having minimal computing costs. Future work should incorporate richer contextual and temporal constraints to improve robustness. [1] Keshri S, Oh M, Zhang S and Iyengar G 2019 Automatic event detection in basketball using HMM with energy based defensive assignment J. Quant. Anal. Sports 15 141–53 [2] McQueen A, Wiens J and Guttag J 2014 Automatically recognizing on-ball screens 2014 MIT sloan sports analytics conference
Read CV Chuqi ChenECSS Paris 2023: OP-AP08
INTRODUCTION: Team handball is a technically-tactically and physically demanding team sport, whereby each playing position require highly specific skills and attributes (Sanchéz et al., 2023). An understanding of the unique demands of each position is vital for developing training programs to optimize adaption processes. Most research aimed at determining the key predictors for different playing positions in handball has traditionally employed confirmatory approaches (Venzke et al., 2024; Sanchéz et al., 2023). The comprehensive data collected from elite team handball opens up opportunities for new exploratory methods, including both unsupervised and supervised machine learning techniques. Thus, our study focuses on the identification of the most important predictors for each playing position in elite team handball by leveraging all available tracking data. METHODS: Data from all teams of the German Bundesliga during the season 2023/24 were obtained from the official data provider (Kinexon, Munich) with the permission of the Dalkin Bundesliga. To increase the validity, a subset was selected based on the following criteria: playing time ≥ 15 minutes, completed pass > 1, matches per player > 9. The final dataset, comprising 301 matches, 249 players, and 142 metrics, was utilized to investigate the demands associated with the six different player positions in team handball. The data set were visually inspected using the t-SNE algorithm and subsequently analyzed with a bootstrap lasso classification (k = 500, n = 249) to reduce bias of repeated measurements. Model evaluation was performed using nested resampling (10-CV). Balanced accuracy was calculated based on the aggregated confusion matrix. The estimated log-odds were used to identify the most important variables. RESULTS: The visual inspection revealed a 4-level categorization for the following playing positions: goalkeepers, wings, backs, and pivots with intersections between wings and backs. Balanced accuracy of the classification was very good (wings: 0.97, pivots: 0.89, backs: 0.93, goalkeepers: 1.00). Highest and lowest log-odds for wings were jumps/min high (beta = 7.7) and passes/min fatal (beta = -4.55), backs sprints/min very high (beta = 28.41), and sprints/min low (beta = -2.36), for pivots passes/min fatal (beta = 3.31) and sprints/min very high (beta = -8.72), and for goalkeepers ball possession/min recovered (beta = 3.01) and accelerations/min medium (beta = -1.00). Additional log-odds help to further clarify the position-specific demands. CONCLUSION: We derived position-specific profiles in elite male team handball from high-dimensional tracking data, revealing patterns that support a four-level classification (goalkeepers, wings, pivots, backs). Extending prior work (García-Sánchez et al., 2023; Venzke et al., 2024), we also incorporated ball-related metrics. Future studies should examine when and how much position-specific training should be integrated into overall training.
Read CV Christian SaalECSS Paris 2023: OP-AP08