************
Introduction
************
This notebook provides the code designed to build and evaluate the super learner models presented in the manuscript "Characterization of high-yield mobility features to identify Parkinson’s disease with a wearable sensor". Each section describes the code of a component of the proposed machine learning pipeline (Fig. 4 in the paper). More details about the pipeline are provided under the 'Methods' section of the paper. The code has been tested using R version 4.1.3 (2022-03-10) and Python version 3.9.13.

************
Setup
************
To view the notebook, open the 'index.html' file in a web browser or open the 'notebook.pdf' file in a PDF viewer application. By following the code in the notebook, the results presented in the paper can be replicated. Also, by loading the final super learner models provided under the 'models' folder and the R objects under the 'rdata' folder of the supplementary material, the same performance measures and feature importance plots can be reproduced. Instructions about how to load an H2O binary model are provided at (https://docs.h2o.ai/h2o/latest-stable/h2o-docs/save-and-load-model.html).

************
Dependencies
************
For each section, the required R packages are listed in the first code chunk. R packages can be installed using the instructions listed in (https://www.dataquest.io/blog/install-package-r/). Only SHAP values calculations are done using Python as described under the 'Model interpretation' subsection of the notebook. The required Python packages are also provided with the code and can be installed as described in (https://packaging.python.org/en/latest/tutorials/installing-packages/).

************
Usage
************
The code reads the sensor data from the 'data' folder which contains two subfolders 'case' and 'control' for cases and controls respectively. Each subject has a separate folder with seven text files (for the 32-feet walk, standing with eyes closed, standing with eyes opened, two trials of TUG, and two trials of CogTUG tasks) and five subfolders to save the segmented components of the complex tasks.