Predicting continuous ground reaction forces from accelerometers during uphill and downhill running: A recurrent neural network solution

Abstract

Ground reaction forces (GRFs) are important for understanding human movement, but their measurement is generally limited to a laboratory. Previous studies used neural networks to predict GRF waveforms during running from wearable device data, but these predictions are limited to the stance phase of level-ground running. We sought to develop a recurrent neural network capable of predicting continuous normal (perpendicular to surface) GRFs across a range of running speeds and slopes from accelerometer data. 19 subjects ran on a force-measuring treadmill at five slopes (0{degrees}, {+/-}5{degrees}, {+/-}10{degrees}) and three speeds (2.5, 3.33, 4.17 m/s) per slope with sacral- and shoe-mounted accelerometers. We then trained a recurrent neural network to predict normal GRF waveforms frame-by-frame. The predicted versus measured GRF waveforms had an average {+/-} SD RMSE of 0.16 {+/-} 0.04 BW and relative RMSE of 6.4 {+/-} 1.5% across all conditions and subjects. The recurrent neural network predicted continuous normal GRF waveforms across a range of running speeds and slopes with greater accuracy than neural networks implemented in previous studies. This approach may facilitate predictions of biomechanical variables outside the laboratory in near real-time and improves the accurately of quantifying and monitoring external loads experienced by the body when running.