Table 4 Feature extraction transforms from 17 source signals to 361 features.
Feature Group | Number of Features | Input Signals | Description |
|---|---|---|---|
Statistical | 102 | Count of magnitude, gross activity, heart rate, time domain heart rate variability, magnitude of Uni counts, A-Fib percent, sleep, respiration rate, step count, tilt, trailing activity, walk percent, skin temperature, heart rate residual, activity residual, respiration rate residual, MCI | The following statistical operations applied to each of the input signals: median, mean, standard deviation, 99th percentile, 1st percentile, interquartile range |
Filtered statistical | 240 | Gross activity, heart rate, time domain heart rate variability, respiration rate, step count*, tilt*, skin temperature, breaths per beat, HRV normalized by HR, heart rate residual, activity residual, respiration rate residual, MCI | Applies the statistical functions median, mean, standard deviation, 99th percentile, 1st percentile to each of the input signals when filtered by the following conditions: (1) while walking, (2) while not walking, (3) while sleeping, (4) while not sleeping. Signals marked with * indicate features only used in conditions 1 and 2. |
Weighted average | 4 | Step count, heart rate, time domain heart rate variability, respiration rate, breaths per beat | Weighted average of input signal using the corresponding step count as the signal weight. |
Interaction | 8 | Step count, heart rate, time domain heart rate variability, respiration rate, breaths per beat | Slope and intercept of linear regression fit to step count against the other input signals, with step count as the dependant variable. Minutes with step count values of zero are removed before fitting the linear regression. |
Delta interaction | 4 | Step count, heart rate, time domain heart rate variability, respiration rate, breaths per beat | Slope of linear regression fit to first order difference of step counts against the first order difference of the other input signals. Minutes with 0 step count first order difference are removed before fitting the linear regression. |
Sleep | 2 | Sleep | Number of awakenings (count of transitions between asleep and awake state), and number of awakenings per hour of sleep. |
Data quality | 1 | ECG Signal quality index | Amount of high quality ECG data per window |
Total features | 361 | Â | Â |
