Author(s): SCHNACK, T., BACA, A., Institution: UNIVERSITY OF VIENNA, Country: AUSTRIA, Abstract-ID: 382

INTRODUCTION:
Heart rate (HR) is an important measure for exercise intensity. While it is increasingly common for football players to wear sensor vests that capture kinematic data (acceleration, angular rate and velocity), these devices do not measure HR. Therefore, the aim of this study is to investigate the use of kinematic data to model a players HR in football. Furthermore, the influence of complexity constraints for model parameter calibration is considered.
METHODS:
Data of five male professional football players was collected during the first half of a friendly game in January 2023. Apex Pro Series (STATSports Group Limited, Newry, GBR) and Firstbeat Sports (Firstbeat Technologies Oy, Jyväskylä, FIN) sensors measured kinematics and HR, respectively. The relationship between kinematics and HR was modelled using linear regression and the Performance Potential model (PerPot) [1]. The data of each player was split in half for calibration and prediction. Parameters of the PerPot were calibrated through a differential evolution algorithm. The complexity of this algorithm was constrained by two values: the maximum number of iterations (default 10) and the population size (default 4). PerPot was calibrated four times, with the default complexity constraints scaled by the factors 1, 4, 7 and 10, respectively. An analysis of variance with repeated measures was used to assess the influence of the model type, the type of kinematic data and the complexity factor on the root mean square error (RMSE) of the prediction.
RESULTS:
PerPot showed a significantly lower RMSE compared to regression (7.6±3.2 vs. 12.8±2.2 bpm, p<0.05). For PerPot, angular rate showed a significantly lower RMSE than velocity (6.8±3.5 vs. 8.23±2.6 bpm, p<0.05) and the complexity factor 1 performed significantly worse than the factors 4, 7 and 10 (11.0±4 vs 7±2.2/6.3±1.8/6.2±1.9 bpm, p<0.05).
CONCLUSION:
While regression only makes the assumption that the modelled relationship is linear, PerPot is based on physiological principles. This might explain the difference in the predictive power of the two models. Velocity is a GPS-derived measure, which might explain its inferiority compared to the inertial-based angular rate. The similarity between complexity factor 4, 7 and 10 suggests that from a certain factor on, higher calibration complexities result in diminishing returns in terms of the accuracy of PerPot. This study proves the feasibility of modelling HR in football, based on kinematic data. However, further research needs to be carried out with more subjects to establish acceptable accuracy limits.
REFERENCES:
[1] Endler, S., Hoffmann, S., Sterzing, B., Simon, P., & Pfeiffer, M. (2017). The PerPot Simulated Anaerobic Threshold—A Comparison to Typical Lactate-based Thresholds. International Journal of Human Movement and Sports Sciences, 5(1), 9–15. https://doi.org/10.13189/saj.2017.050102