EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Machine learning-based prediction of head impact dynamics using an instrumented cycling helmet

Ben Emsley, Jon Farmer, Paul Goodall, Paul Sherratt, Tom Jackson, Andrew West

Intelligent Sports and Health · 2025

DOI: 10.1016/j.ish.2025.06.002 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Intelligent Sports and Health, 2025. Verfasst von Ben Emsley und 5 weiteren Autor:innen. Die Arbeit nennt 76 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Real-time head impact data are crucial for advancing cyclist safety; however, current helmet-mounted devices are frequently limited by substantial measurement errors. This research explores a machine learning (ML) approach to enhance the interpretability of sensor data, with the aim of classifying impact locations and predicting headform peak linear acceleration (PLA). Two ML algorithms, random forest and extreme gradient boosting (XGBoost), were evaluated using data from 101 impacts (45 front-boss and 56 side impacts) on a conventional road cycling helmet instrumented with three ADXL373 accelerometers. From each sensor, 194 time- and frequency-domain features were extracted and refined through dependency removal, SHapley Additive exPlanations (SHAP) importance rankings, and exhaustive feature selection. The models were trained using both multisensor and single-sensor data for comparison. The multisensor classification models achieved high accuracy and area under the receiver operating characteristic (AUROC) curve scores (both 1.00) on independent test data, while the single-sensor models yielded good but slightly lower scores (0.90 to 0.97). For PLA prediction, the XGBoost multisensor models demonstrated the highest performance, with R 2 values of 0.82 ± 0.06 and 0.82 ± 0.08 and MAEs of 7.1 ± 1.5 g (7.3 ± 1.3 %) and 7.2 ± 1.2 g (8.3 ± 2.2 %) for front-boss and side impacts, respectively. SHAP analysis revealed that frequency domain features were the most influential predictors. These findings highlight the potential of ML models, the advantages of using a multisensor system and the importance of frequency-domain features for characterising head impacts in cycling, which may offer valuable insights into the development of smart helmet technology and improved cyclist safety. • Machine learning algorithms are used to characterise head impacts using embedded helmet sensors. • Models trained with multisensor data outperform those trained with single-sensor data. • Frequency-domain features are the most influential for model performance.

Zitieren

APA

Emsley, B., Farmer, J., Goodall, P., Sherratt, P., Jackson, T., & West, A. (2025). Machine learning-based prediction of head impact dynamics using an instrumented cycling helmet. Intelligent Sports and Health. https://doi.org/10.1016/j.ish.2025.06.002

Vancouver

Emsley B, Farmer J, Goodall P, Sherratt P, Jackson T, West A. Machine learning-based prediction of head impact dynamics using an instrumented cycling helmet. Intelligent Sports and Health. 2025. doi:10.1016/j.ish.2025.06.002

BibTeX
@article{Emsley2025,
  author = {Emsley, Ben and Farmer, Jon and Goodall, Paul and Sherratt, Paul and Jackson, Tom and West, Andrew},
  title = {Machine learning-based prediction of head impact dynamics using an instrumented cycling helmet},
  journal = {Intelligent Sports and Health},
  year = {2025},
  doi = {10.1016/j.ish.2025.06.002}
}
Kostenlos starten →