Predicting vertical ground reaction force in rearfoot running: A wavelet neural network model and factor loading

Area of Science:

• Biomechanics
• Computational modeling
• Sports science

Background:

• Vertical ground reaction force (vGRF) is crucial for understanding running mechanics.
• Accurate prediction of vGRF can aid in injury prevention and performance optimization.
• Previous methods for vGRF prediction often involve complex feature selection.

Purpose of the Study:

• To propose a simple yet effective method for selecting input variables for vGRF prediction.
• To utilize factor loading for kinematic variable selection in a wavelet neural network (WNN) model.
• To evaluate the accuracy and effectiveness of the proposed WNN model for vGRF prediction across different running speeds.

Main Methods:

• Collected kinematic data and vGRF from 9 rearfoot strikers at 12, 14, and 16 km/h using motion capture and an instrumented treadmill.
• Employed factor loading to screen and select significant kinematic input variables.
• Developed and validated a WNN model using the selected variables to predict vGRF.

Main Results:

• Factor loading identified 9 key kinematic variables, primarily related to knee and ankle joints.
• The WNN model demonstrated high prediction accuracy for vGRF (CMC > 0.98, NRMSE < 15%) across all tested speeds.
• Specific components like impact force, active force, and peak time were predicted with high accuracy (NRMSE < 15%).

Conclusions:

• Factor loading is a valid and efficient method for selecting kinematic variables for artificial neural network models.
• The WNN model effectively predicts vGRF, with the knee joint being the optimal predictor.
• The findings support the use of this method for accurate vGRF estimation in running analysis.