Calculate the Activity Index (AI)

Computes the Activity Index of Bai et al. (2016) from raw tri-axial accelerometry. For each 1-second window \(t\), with \(\sigma^2_m(t)\) the variance of axis \(m \in \{X, Y, Z\}\) over the samples in that second, $$AI(t) = \sqrt{\frac{1}{3}\sum_{m} \sigma^2_m(t)}.$$ The per-second values are then summed within each unit (e.g. a minute).

Usage

acti_calculate_ai(data, unit = "1 min", ensure_all_time = TRUE, verbose = TRUE)

acti_calculate_activity_index(
  data,
  unit = "1 min",
  ensure_all_time = TRUE,
  verbose = TRUE
)

Arguments

data

An object with columns X, Y, and Z and a time column, or an object of class AccData

unit

length of time to sum the per-second AI over, passed to lubridate::floor_date()

ensure_all_time

if TRUE, all times from the first to the last unit are in the output, even those with no input data

verbose

print diagnostic messages

Value

A data.frame (or data.table if data is one) with columns HEADER_TIME_STAMP and AI.

Details

This is the unadjusted index (sometimes written \(AI_0\)): it does not subtract or normalise by the device's systematic noise variance \(\bar{\sigma}^2_m\), which the original definition estimates from a stationary period. It differs from AI_DEFINED returned by acti_calculate_mad(), which takes the variances over the whole unit rather than summing 1-second values.

A 1-second variance needs at least two samples, so seconds with fewer than two samples are NA and are skipped when summing; a unit with no valid seconds is NA. Data sampled at under 2 Hz therefore give all-NA AI (with a warning).

References

Bai J, Di C, Xiao L, et al. (2016). An Activity Index for Raw Accelerometry Data and Its Comparison with Other Activity Metrics. PLoS ONE 11(8): e0160644. doi:10.1371/journal.pone.0160644

Examples

file = actiread::acti_example_gt3x()
res = actiread::acti_read_gt3x(file, verbose = FALSE)
ai = acti_calculate_ai(res[1:12000, ])
#> Running floor_sec on time
#> Summarizing the variance
#> Calculating AI