Simulation of interaction patterns and simulative effectiveness analysis of creative plays in team sports by means of neural networks
Final Report Abstract
The description and analysis of complex group interactions, as they occur in team sports such as football, can often not be reduced to static distributions of numbers, if the interrelationships are to be examined comprehensively and the underlying dynamics of interaction processes are to be recorded. Therefore, the success of specific actions cannot, for example, simply be measured by means of percentages without placing the actions in context of the situations in which they took place. This question was examined in context of the present project, based on the concrete example of match analysis in high-performance football. Based on the concepts developed in previous projects, a computer-aided analysis procedure was developed which recognizes process patterns of the game on a basis of position data and checks their effectiveness, too. The focus was, in particular, on the identification of creative processes. Creative processes were those actions which occur rarely and are successful at the same time. The concept of action can very broadly be understood in football from the technical individual action, over a group tactical action, to the complex process of attack or defense. Based on the specifically investigated group-tactical structures of attack and defense formations, it was obvious to use the choice of an attack formation in context of the opponent’s defense formation and vice versa as representatives of action. For this purpose, position data, which is the localization of the players during the game, were first transferred into discrete tactical pattern categories with the help of a neural network. This allowed the quantification of the rarity of certain tactical patterns by comparing the coincidences of pattern pairs (attack-defense). Subsequently, the success of the action was determined. While in team games such as handball, basketball or volleyball success of a move can be measured as a goal, basket or point, the goal as success of a move in football is very rarely and only very weakly determined by the play (about 2.7 goals/game). Instead, success indicators such as space gain or the number of opponents passed by the ball can be used, the improvement of which can be seen as the success of an action such as a pass. Within the framework of this project, a large-scale study was conducted to verify the validity of the developed success indicators. A correlative study based on position data from 103 Bundesliga matches showed that both indicators (number of opponents passed by the ball and changes in space gain) are significantly related to the game's success. Based on combination of tactical patterns, their frequencies and the resulting success, a simulation was carried out using a state-event model (Z-E-model). Here it could be shown which state transitions result in successful or less successful constellations. The present project showed how artificial neural networks can be used to map complex group interactions based in the team sports football. The results, especially in the area of success indicators, enable an automated evaluation of game actions, which then can be examined for improvement possibilities by means of simulation.
Publications
- (2013). Neural Networks for Analysing Sports Games. In T. McGarry, P. O’Donoghue, & J. Sampaio (Eds.), Routledge Handbook of Sports Performance Analysis (pp. 237-247). Abingdon: Routledge
Perl, J., Tilp, M., Baca, A., & Memmert, D.
- (2013). Tactical Creativity. In T. McGarry, P. O’Donoghue, & J. Sampaio (Eds.), Routledge Handbook of Sports Performance Analysis (pp. 297-308). Abingdon: Routledge
Memmert, D.
- (2013). Tactics in soccer: an advanced approach. International Journal of Computer Science in Sport, 12, 33-44
Perl, J., Grunz, A., & Memmert, D.
- (2014). Analysis of process dynamics in soccer by means of artificial neural networks and Voronoi-cells. In A. Baca (Ed.). Sportinformatik 2014 (11. Symposium der Sektion Sportinformatik der Deutschen Vereinigung für Sportwissenschaft vom 12.-14. Sept. 2014 in Wien)
Perl, J., & Memmert, D.
- (2014). Possession vs. Direct Play: Evaluating Tactical Behavior in Elite Soccer, International Journal of Sports Science. 4(6A), 35–41
Kempe, M., Vogelbein, M., Memmert, D., &, Nopp, S.
(See online at https://dx.doi.org/10.5923/s.sports.201401.05) - (2015). Detecting tactical patterns in basketball: comparison of merge self-organising maps and dynamic controlled neural networks. European Journal of Sport Science, 15(4), 249-255
Kempe, M., Grunz, A., & Memmert, D.
(See online at https://doi.org/10.1080/17461391.2014.933882) - (2016). Big data and tactical analysis in elite soccer: future challenges and opportunities for sports science. Springerplus, 5(1), 1410
Rein, R., & Memmert, D.
(See online at https://doi.org/10.1186/s40064-016-3108-2) - (2016). Current Approaches to Tactical Performance Analyses in Soccer using Position Data. Sports Medicine
Memmert, D., Lemmink, K. A. P. M., & Sampaio, J.
(See online at https://doi.org/10.1007/s40279-016-0562-5) - (2016). Evaluation of changes in space control due to passing behavior in elite soccer using Voronoi-cells. In P. Chung et al. (Eds.), Proceedings of the 10th International Symposium on Computer Science in Sports (ISCSS) (Vol. 392, pp. 179-183). Springer International Publishing
Rein, R., Raabe, D., Perl, J., & Memmert, D.
(See online at https://doi.org/10.1007/978-3-319-24560-7_23) - (2016). Key Information From Complex Interaction Processes in Football. Research Quarterly for Exercise and Sport, 87:sup1, S72
Perl, J., & Memmert, D.
(See online at https://dx.doi.org/10.1080/02701367.2016.1200446) - (2016). Soccer analyses by means of artificial neural networks, automatic pass recognition and Voronoicells: An approach of measuring tactical success. In P. Chung et al. (Eds.), Proceedings of the 10th International Symposium on Computer Science in Sports (ISCSS) (Vol. 392, pp. 77-84). Springer International Publishing
Perl, J., & Memmert, D.
(See online at https://doi.org/10.1007/978-3-319-24560-7_10) - (2016). Spatio-Temporal Convolution Kernels. Machine Learning, 102(2), 247-273
Brefeld, U., Knauf, K., & Memmert, D.
(See online at https://dx.doi.org/10.1007/s10994-015-5520-1)
