Figure 2.

Combinations of commonly used data preprocessing steps before machine-learning classifications. (1) Data points per foot and dimension. (2) Time-continuous waveforms without reduction (TC), time-discrete gait variables by an unsupervised reduction (TD), and principle components by a supervised reduction using Principal Component Analysis (PCA). (3) Z-transformation combined with scaling from [−1, 1] over single trials (ST) or all trials (AT). f_c: individual optimal filter cut-off frequency. (4) ΔtGRF: first-time derivative of GRF.