Offwrist-detection

This program can label the data from actigraphy as non-wear or wear based on a trained random forest algorithm

You can test the program with the actigraphy series inside ./data/ directory.

The real label is presented on column "NA2":

• 0: Wearing

• 1: Not wearing

Requirements

You need to install Python 3.8+ to run this code. Check if Python is installed by using the following command:

> python --version

Output:

Python 3.8.x

To install all the needed packages to run this code. In the folder of this project, run the following command:

> pip install -r requirements.txt

Running the code

You can run the program with the following command:

> python main.py "actigraphy_sequence.xlsx"

The output will be added to the end of your xlsx file. Output example:

........	timeVar	NA2	NA3	ML_OffWrist_Prediction
........	28/09/2020 00:00	1	NA	1
........	28/09/2020 00:01	1	NA	1
........	28/09/2020 00:02	1	NA	1

Datasets

The data are divided in three directories:

• ./data_train/ : off-wrist periods shorter than 30 minutes were considered wear for training purposes (true label = "NA2" column).
• ./data_raw/ : raw data from the actimeters - HA was run on those in our publication.
• ./data_test/: The column "NA2" is the user record for off-wrist (with intervals <30min included; true label). The results of all performances reported in our publication were computed using this. All algorithms except HA were run in these data.

Data from our validation (proof-of-concept) cannot be made available, but results are described in the publication.

Try the code using our examples in ./data/

> python main.py "data/01.xlsx"

HA function

You can also find the HA function here.

*We used v3.1 (filterAround = T, filterBet = T) in the publication analyses, but no important change in results is seen with fixes/edits.