The aim of this study was to investigate the efficacy of common machine learning algorithmic approaches to estimate lower limb joint moments during fast walking gait. Kinematic and ground reaction force data on 19 participants were captured with a force-plate and motion caption capture system. Inverse dynamics was used to calculate the right lower limb joint moments and common machine learning algorithmic approaches, such as Random Forest (RF), Linear Regression (LR), Neural Network (NN), AdaBoost (AB) and Gradient Boosting, were used to predict the corresponding joint moments using only the kinematic data. High coefficient of determination values (R2>0.9) for predicting moments using random forest, neural network and AdaBoost are observed in for the ankle, knee and hip joints in frontal, sagittal and transverse planes. The other approaches had R2 values between ranged 0.71 and 0.97. This suggests that common machine learning algorithms may be a feasible approach to estimate joint moments during fast walking in a clinical setting for monitoring sport injury prevention and management.
Ong, Alex Dr and Hamill, Joseph
"ESTIMATING LOWER LIMB JOINT MOMENTS IN GAIT USING COMMON MACHINE LEARNING APPROACHES,"
ISBS Proceedings Archive: Vol. 41:
1, Article 87.
Available at: https://commons.nmu.edu/isbs/vol41/iss1/87