Table 3. Error metrics for the predicted waveforms and discrete variables when training the long short-term memory (LSTM) network with and without foot strike pattern as an input feature.

Root mean square error (RMSE) and relative RMSE (rRMSE) are presented for predicted normal ground reaction force (GRF) waveforms. Mean absolute percent error (MAPE) values are presented for the discrete variables calculated from normal GRF waveforms predicted by both LSTM networks.

	LSTM with foot strike (Sacral + Right foot accelerometers)	LSTM without foot strike (Only sacral accelerometer)
GRF waveform
RMSE [BW]	0.16 ± 0.04	0.17 ± 0.05
rRMSE	6.4 ± 1.5%	6.7 ± 1.7%
Discrete variables MAPE
Step frequency	0.1 ± 0.1%	0.1 ± 0.1%
Contact time	4.9 ± 4.0%	5.6 ± 4.5%
Impulse	6.4 ± 6.9%	6.0 ± 7.1%
Active peak	8.5 ± 8.2%	7.7 ± 6.3%
Loading rate	27.6 ± 36.1%	30.3 ± 41.6%