Skip to content

Latest commit

 

History

History
36 lines (25 loc) · 2.01 KB

README.md

File metadata and controls

36 lines (25 loc) · 2.01 KB
Raw

Actometer behavioral scoring from wrist- or hip-worn Actigraph data

To apply previously trained model on new data:

  1. Use ActiLife software to export the raw Actigraph data to .gt3x format. Skip this step you are an accelerometer brand that is not Actigraph.
  2. Install R and RStudio. If they are installed, check that they are the latest version.
  3. Store ActChronicFatigue.R on your computer.
  4. Open ActChronicFatigue.R in RStudio.
  5. Click the [Source] button in RStudio.
  6. Follow the instructions in the console.
  • The first time you do all software will be installed, which may take a while. The second time you do this you will be asked (in Dutch) whether you want to install the software again. Select No.
  • Next, the software will ask you to specify the locations of your data.
  • Once that is done the software will continue with processing the data, and classifying the data.
  • A summary of the findings will be printed to the screen and also stored in the output directory.

Note: The software assumes that the participant ID is stored in the filename before the first space.

To train a new model based on existing training data:

  1. Follow steps above
  2. Create labels.csv file with one column for id, one column for label (holding character values for "pp" and "fa") and one column loc specifying the body location ("wrist" and "hip").
  3. Run script fitmodel.R after updating the info at the top to match your situation.

Model interpretation

The models are logistic regression models, which can be interpretted as follows:

If the coefficients are 4.75861402 and -0.05916747, then
x = 4.75861402 + (-0.05916747 * act9167 )
probability_pp = 1/(1+ exp(-x))

In other words: a lower value of act9167 (more inactive person) will result in higher x, which will increase the probability of being pp (pervasively passive), while a higher value of act9167 (more active person) will result in a lower value of x and result in a lower probability of being pp (pervasively passive).