While testing of the deep learning model on a balanced dataset resulted in a very promising classification performance (91.5 % sensitivity, 94.3 % precision), the evaluation on an unbalanced dataset (great amount of non-header events vs. small amount of true header events), as typically obtained in real-life settings, revealed a precision score of only 42.2 %, which would result in an overestimation of soccer players’ true heading exposure. However, when compared to other available methods for automatically detecting soccer headers from wearable sensor data, our neural network still achieved superior classification scores and, therefore, can be seen as a valuable step towards an effective as well as efficient quantification of soccer players’ individual heading exposure.
Kern, J., Lober, T., Hermsdörfer, J., & Endo, S. (2022). A neural network for the detection of soccer headers from wearable sensor data. Scientific Reports (12). doi.org/10.1038/s41598-022-22996-2