Data Collection - Creating Scenarios

[hooman007]

Think and Plan for data collection experiments in different scenarios

We want to plan for some experiments and come up with procedures to collect and label data. Use Trier Social Stress procedure and papers as guide.

Requirements and Subtasks:

• Procedures must include:
  • what the demographics of users are
  • exact steps of experiments
  • which features of fitbit are main targets of data collection
• Document the procedures on google doc.

NOTE: Please Consider the points bellow in making scenarios.

Feedbacks from Jack (? daniel's psych roommate):

• Limit Demographics! Focus on one demographics for the initial database to limit noise and errors eg. females with age 20-28
• Get a baseline, sitting, doing nothing. Then, get an active state which is Not Stress, like Motion Object Tracking task, Mentally Active but not stressed.
  • Trier Social Stress task:
    • echologically valid test. with a couple of steps.
    • we can simplify it, but the simpler, the less valid
    • get the most recent publication that uses this protocol and uses the data we want. then we can use that.
    • ask profs as well.

Fitbit related and Technical aspects

• Consider the capability of fitbit (features it can track) and possible help each feature can provide us in the project, to make scenarios for data collection experiments. The features can be found here: Ticket #2 Google Sheet: https://docs.google.com/spreadsheets/d/1RTC4KSgaOXiYIjxeeEIBpQEpontjNQ1WJbvkz-ECCtM/edit?usp=sharing

• Consider the data collection interface python script : Ticket #3

• ask for user's feedback on how much they are stressed to label the data. e.g after each section of procedure that we collect data, ask them to rate their stress 0-10 and give us reasons maybe. because a scenario might be more scary to someone that someone else.

[hooman007]

data collection is now available. you can start the experimentations.

[hooman007]

Based on the recent talks, we decided to go to 2 directions.

1. have a short term prediciton model which is sensitive to spikes of HR and GSR (galvanic skin resistance) to detect sudden anxiety or just short term issues (like in order of hours or minutes)

2. have a long term model for mood and anxiety in long term whcih uses other inputs, mainly acquired input from user itself. we chose 3 categories for now, mood, anxiety, energy level.

For now, we willl focus on the first model (the short term) and get only the HR as we use FITBIT. unless GSR is ready from HW team.

So design experiments that will get HR and GSR as inputs and ANXIETY level as output for different situations that take several minutes. But it should be long enough to match the fitbit resolution.

Note: check the HR resolution of FITBIT. currently the smalles resolution is 5 seconds based ont he CSV of data of Jan 10. FITBIT says they can have resolution of 1 second. So maybe try to have quick test in which you:

• use the fitbit and change your heart rate quickly (like deep breathing and then jumping up and down).
• use the data extraction script to get the HR data of the same day,
• check the resolution of the given data.