This study presents a non-invasive approach using machine learning to predict proximal tibial anterior shear force (ASF), a surrogate for anterior cruciate ligament (ACL) loads from inertial measurement unit (IMU) data, providing a practical alternative to direct force measurement. Employing XGBoost, the research analysed IMU data from drop jump tasks performed by 22 female participants and validated on an additional participant. The model underwent optimization through feature reduction and signal filtering. The results demonstrate an XGBoost model using IMU data to estimate ASF showed improved prediction accuracy after feature reduction and a low-pass filter. The model was able to predict ASF with a root mean squared error of 41.59±13.18N, a mean absolute error of 35.18±11.97N, and an R² value of 0.84±0.07.
© Copyright 2024 ISBS Proceedings Archive: Vol. 42: Iss. 1. All rights reserved.