A comparison of three methods for estimating vertical ground reaction forces in running

Abstract

The purpose of this study was to compare different approaches for the estimation of biomechanical loads in running. A neural network, a biomechanical model, and a two-mass model were tested on the same data set. The predictions of the neural network were highly accurate for all considered running speeds (average RMSE, 0.11 BW). The biomechanical model returned statistically similar results (p=0.113, 0.14 BW), but with increasing RMS errors at high running speeds. Finally, the two-mass model estimates were independent of running speed, but were the least accurate (RMSE, 0.18 BW).