DEVELOPMENT OF A NOVEL ALGORITHM FOR THE AUTOMATIC DETECTION OF MULTIDIRECTIONAL LOCOMOTION WITHIN TEAM SPORTS

Field-based sports are characterised by their intermittent nature requiring both, multidirectional locomotion and, sports-specific movements at a range of intensities. Traditionally, athlete-monitoring has focused on quantifying workload based on movement intensity with minimal regard to the direction of locomotion [1]. The aim of this study was to develop and evaluate an algorithm to detect and classify multidirectional movement using signal characteristics from a microtechnology device. Rugby league referees (n=13) undertook a match-play simulation protocol (i.e., changes in movement speed and locomotion direction) [2], with microtechnology and video data collected across five-trials. Video data was reviewed to identify movement anomalies outside of the simulation protocol for exclusion. From the 100Hz microtechnology data, acceleration measures were used to classify the start and end point of each movement (i.e., backwards, forwards, sideways or other) or marked for exclusion from the algorithm development. The classified sensor data was processed in Python (v3.11), where data were split into training and testing datasets. A Recurrent Neural Network (Long Short-Term Memory) [3] was implemented to develop and validate an algorithm. Model performance was assessed via accuracy, sensitivity, precision and Area Under the Receiver Operating Characteristic Curve (AUC), using the testing dataset. The accuracy of the model was 0.973 ± 0.010. Sensitivity and precision of the model varied between movement direction, but was >0.928 and >0.922, respectively. The AUC of the model was 0.988 ± 0.007. The current study highlights the effectiveness of a microtechnology based algorithm for automatically classifying multidirectional locomotion of various velocities. Practically, such algorithm can be used to inform evidence-based training in relation to multidirectional locomotion. Whilst model performance was very-high, further research should examine the feasibility of applying the algorithm to match-play datasets to enhance athlete-monitoring processes.