actimetric

The goal of actimetric is to apply machine learning-based algorithms to accelerometer data to provide insights in physical activity and sleep behaviors in free-living conditions.

Installation

To use actimetric, first you need to install the development version of actimetricModels like this:

library(remotes)
install_github("PhysicalActivityOpenTools/actimetricModels")

Then, you can install the development version of actimetric like so:

library(remotes)
install_github("PhysicalActivityOpenTools/actimetric")

Example usage

To use actimetric, you can just rely on the runActimetric function.

The bare minimum of input arguments would be the input_directory, output_directory, and the classifier that you would like to use. See next function call for a example:

library(actimetric)

# This runs the program
runActimetric(input_directory = "G:/directory_containing/the_input_files/",
              output_directory = "G:/directory_to_save/the_output_files/",
              studyname = "my_study_name",
              classifier = "Preschool Wrist Random Forest Free Living")

The actimetric R package supports the use of 9 different classifiers for now:

• Preschool Wrist Random Forest Free Living

• Preschool Hip Random Forest Free Living

• Preschool Hip Random Forest Free Living Lag-Lead

• Preschool Wrist Random Forest Free Living Lag-Lead

• School age Wrist Random Forest

• School age Hip Random Forest

• Adult Wrist RF Trost

• Adult women Wrist RF Ellis

• Adult women Hip RF Ellis

Additional parameters

Additionally, you can control a number of extra parameters in actimetric. For example, you can decide on calibrating the data with argument do.calibration, on the acceleration metrics you would like to calculate with options for ENMO, ActiLife Counts, and ActiLife LFE counts, or whether to detect non-wear and sleep periods.

Here a function call with all potential arguments you may use in actimetric, all of them defined with their default values:

library(actimetric)

# This runs the program
runActimetric(input_directory = NULL,
              output_directory = NULL,
              studyname = "actimetric",
              classifier = NULL,
              do.calibration = TRUE,
              do.sleep = TRUE,
              do.nonwear = TRUE,
              do.enmo = FALSE,
              do.actilifecounts = TRUE,
              do.actilifecountsLFE = FALSE,
              boutdur = 10, boutcriter = 0.8,
              verbose = TRUE, overwrite = TRUE,
              n_valid_hours = 16,
              n_valid_hours_awake = 10,
              n_valid_hours_nighttime = 2,
              visualreport = TRUE)

The output

The actimetric R package will generate two folders inside the output_directory. One of them will contain the time series with the time-stamped epoch-level estimates for the activity classification and the acceleration metrics calculated. The time series come in the format of .RData. Additionally, actimetric will generate another folder, named results, that will contain day-level and recording-level aggregates of the data in .csv format, as well as visualizations of the recordings in .pdf format. Note that visualizations are not generated by default, yet you can turn them on with argument visualreport.