Introduction For the identification of game tactics and the selection of successful playing strategies analysis of game situations is very important. Previous own studies focused on identification of offensive playing patterns in team handball (Schrapf & Tilp, 2013). In order to get deeper insight into team and player tactics, the behavior of the defensive teams also has to be considered. Therefore, the aim of the present study was to classify defense situations. Methods For the present study 12 games from the EHF EURO Men 18 in Hard (Austria) were captured by 8 cameras. Subsequently, shot-actions were annotated with custom-made software. Every annotation includes the ground position of the player performing the shot and of all defensive players at the instant of the shot. In total, 728 actions were annotated which were then analyzed by artificial neural network software (Perl, 2002). In order to obtain suitable entropy, data was enlarged by multiplication to a quantity of 7280 datasets with a noise of 5% and subsequently permutated to minimize unwanted learning effects due to duplication. Position data of the shot and the defensive players were used to train the neural network with a dimension of 400 neurons. Each neuron represents a pattern of defense action. Hereafter, similar neurons are grouped to clusters which represent similar defense behavior. The similarity resolution, which defines selectivity between similar and dissimilar neurons, was set to 75%. Results The artificial neural network recognized 18 clusters and 3 single neurons which could not be assigned to a cluster. Thus, we found 21 different patterns of defense positions. The different defense patterns coincide very well with the usual used shot position areas in team handball. The network determined two different patterns for the right wing and left back position, three from the left wing, the right back, and the pivot position, and five for shots from the center back. Discussion The analysis revealed the applicability of artificial neural networks for identifying defense patterns in team handball. As expected, the orientation of the defense coincides with the position of the shot. Differences between the single defense patterns mainly consist in the distance to the goal and the width and orientation (center, right, or left) of the defense. Furthermore, the neural network determined five defensive patterns where one player takes over an offensive role. Further analysis including action sequences preceding the shot positions instead of considering only the single shot position have to be done in order to observe the team tactics in its entirety.
© Copyright 2014 19th Annual Congress of the European College of Sport Science (ECSS), Amsterdam, 2. - 5. July 2014. Veröffentlicht von VU University Amsterdam. Alle Rechte vorbehalten. / Übersetzt mit https://www.DeepL.com/Translator