Classification of spatio-temporal data represented as multiset for tactical profiling in badminton

In the high pressure environment of international sports, tactical advantages are gained through proficient anticipation of opponent movement patterns and the efficient execution of counter movements. Such anticipation and decision making is reliant on the comprehension of past tactical knowledge. We have proposed a method to classify and visualise tactical movement data in badminton, allowing the generation of tactical profiles and efficient comprehension of prior tactical knowledge. We have implemented an automated method to gather player trajectory data from a badminton match, and to annotate these trajectories on tactically important cells on a badminton court. Tactically important movement groups were defined, and the trajectories classified according to their tactical relevance. The classification approach performs at 93.38% accuracy, 93.22% precision and 92.83% recall. We have also proposed a method to eliminate the need to train a classifier, where a representative trajectory for each tactical movement group was defined and the player trajectories grouped based on their similarity to the respective representative trajectory. The proposed grouping approach performs better than the classification at 96.32% accuracy, 95.93% precision and 95.76% recall.