Algorithm 1: Steps for pre-processing the raw accelerometry data.

Input: $x\left(t\right)$—accelerometry signal, $f_s$—sampling frequency, $s_{min}=1.2$ Hz, $s_{max}=4.0$ Hz  Output: $FFT$—scaled average FFT spectrum, $VMC$—average VMC, $Cadence$—average cadence 1 Divide accelerometry signal into 10 s non-overlapping windows. 2 Transform accelerometry signal into vector magnitude $vm\left(t\right)$. 3 Compute vector magnitude count, $v_{10}\left(t\right)$, for each window. 4 Compute Fourier spectrum for each window. 5 Estimate cadence as frequency centered under the largest peak in spectrum. 6 Transform spectra from frequency domain to order domain by scaling frequency axis by the frequency of the cadence. 7 Average vector magnitude count, cadence, and order domain spectra across all windows. 8 Restrict spectra to points between 0.3 and 5.75 multiples of the cadence frequency. 9 Scale spectra by magnitude of the average spectra at the cadence.