Table 2. Continuous Features from Sensor Records via Participant Accelerometers.

ENMOtrunc	Euclidean Norm Minus One truncated (make zero if negative)
ENMOabs	Euclidean Norm Minus One absolute (make positive if negative)
Mean	mean of vector magnitude (+x/y/z axis)
Sd	standard deviation of vector magnitude (+x/y/z axis)
RMS	Root Mean Square (quadratic mean, average power) (+x/y/z axis)
TAC	Total Activity Count (total motion amount) (+x/y/z axis)
MCR	Mean Crossing Rate (across signal average) (+x/y/z axis)
MMCR	Average of Min-Max Crossing Rate (+x/y/z axis)
MAD / MPD	Mean Amplitude Deviation / Mean Power Deviation
Min / Max	Minimum / Maximum vector magnitude (+x/y/z axis) 25thp/Median/75thp
Range	total range of x/y/z axis
kurt/skew	kurtosis / skewness (shape of distribution)
CoV	CoVariance in xy/xz/yz axis
corr	Correlation in xy/xz/yz axis
autocorr	autocorrelation of acceleration (time series) (+x/y/z axis)
coefvariation	coefficient of variation (relative dispersion) (+x/y/z axis)
pitchg/rollg/yawg	magnitude (gravity) of accelerometer rotation (3D)
sdpitch/sdroll/sdyaw	standard deviation of accelerometer sensor rotation
avgpitch/avgroll/avgyaw	average of accelerometer sensor rotation

76 Variables for Sensor Features: Accelerometer Derived in Time Domain

38 computed using Biobank software on field 90001

38 further derived using Biobank frequency domain

Biobank software: ENMOtrunc, Mean/Sd+x/y/z (9)

MAD/MPD, kurt/skew (4)

Min/Max, 25thp/Median/75thp, autocorr/coefvariation (7)

Range x/y/z, Cov yz/xz/xy, corr yz/xz/xy (9)

pitch/roll/yaw g/sd/avg (9)

Derived Software: ENMOabs (1)

x/y/z for Min/Max, 25thp/Median/75thp, autocorr/coefvariation (21)

new features computed: RMS/TAC/MCR/MMCR +x/y/z (16)