Skip to contents

Estimate Non-wear from Raw Accelerometry with the actipy Method

Source: R/acti_calculate_wear_raw_actipy.R
acti_calculate_wear_raw_actipy.Rd

Detects non-wear from raw tri-axial acceleration with the stationary-episode definition of Doherty et al. (2017), as implemented in actipy's detect_nonwear() (Chan and Doherty). This is a standalone port and does not require actipy or Python. Usually called through acti_calculate_wear_raw().

Usage

acti_calculate_wear_raw_actipy(
  data,
  window = 10,
  patience = 5400,
  sd_threshold = 0.015
)

Arguments

data

A data.frame of raw accelerometer data with a time column and X, Y, Z columns, or an AccData object. Data should be calibrated (e.g. with acti_calibrate()).

window

Length, in seconds, of the windows used to test for stationarity.

patience

Minimum span, in seconds, of a stationary run for it to be non-wear.

sd_threshold

Standard deviation (g) below which an axis is considered still.

Value

A tibble with one row per window: time (start of the window), stationary, nonwear, and wear.

Details

The signal is split into windows of window seconds, anchored at the first sample. A window is stationary when the standard deviation of every axis is below sd_threshold. A run of stationary windows is non-wear when its span, from the start of its first window to the start of its last window, is longer than patience seconds. Windows with fewer than two samples (for example, during time gaps) are not stationary.

The defaults are actipy's (10 seconds, 90 minutes, 15 mg). Doherty et al. (2017) used 60 minutes and 13 mg (patience = 3600, sd_threshold = 0.013). actipy sets samples to NaN from the start of the first window to the start of the last window of a run; this function labels whole windows, so the last window of a run is labelled non-wear. actipy runs this step after low-pass filtering and gravity calibration.

References

Doherty A, et al. (2017). Large scale population assessment of physical activity using wrist worn accelerometers: the UK Biobank study. PLoS ONE 12(2): e0169649. doi:10.1371/journal.pone.0169649

Chan S, Doherty A. actipy: a Python package to process wearable sensor data. doi:10.5281/zenodo.14514488

See also

acti_calculate_wear_raw(), acti_calculate_wear_raw_ggir()

Examples

n = 3 * 3600 * 10
time = as.POSIXct("2020-01-01 00:00:00", tz = "UTC") + (seq_len(n) - 1) / 10
moving = rep(c(TRUE, FALSE, TRUE), each = n / 3)
set.seed(1)
data = data.frame(
  time = time,
  X = ifelse(moving, stats::rnorm(n, sd = 0.3), 0),
  Y = ifelse(moving, stats::rnorm(n, sd = 0.3), 0),
  Z = ifelse(moving, stats::rnorm(n, sd = 0.3), 1)
)
wear = acti_calculate_wear_raw_actipy(data, patience = 3000)
table(wear$wear)
#> 
#> FALSE  TRUE 
#>   360   720