data

What do you think about an app designed for student’s smartphones to track mental health so that students and college officials can, in the future, keep mental health issues from interfering with learning?  The mining of human behavior for 48 Dartmouth students is described below from dartmouth.edu.

 

StudentLife is the first study that uses passive and automatic sensing data from the phones of a class of 48 Dartmouth students over a 10 week term to assess their mental health (e.g., depression, loneliness, stress), academic performance (grades across all their classes, term GPA and cumulative GPA) and behavioral trends (e.g., how stress, sleep, visits to the gym, etc. change in response to college workload — i.e., assignments, midterms, finals — as the term progresses).

 

Much of the stress and strain of student life remains hidden. In reality faculty, student deans, clinicians know little about their students outside of the classroom. Students might know about their own circumstances and patterns but know little about classmates. To shine a light on student life we develop the first of a kind StudentLife smartphone app and sensing system to automatically infer human behavior. Why do some students do better than others? Under similar conditions, why do some individuals excel while others fail? Why do students burnout, drop classes, even drop out of college? What is the impact of stress, mood, workload, sociability, sleep and mental health on academic performance (i.e., GPA)? The study used an android app we developed for smartphones carried by 48 students over a 10 week term to find answers to some of these pressing questions.

We use computational methods and machine learning algorithms on the phone to assess sensor data and make higher level inferences (i.e., sleep, sociability, activity, etc.) The StudentLife app that ran on students’ phones automatically measured the following human behaviors 24/7 without any user interaction:

  • bed time, wake up time and sleep duration
  • the number of conversations and duration of each conversation per day
  • physical activity (walking, sitting, running, standing)
  • where they were located and who long they stayed there (i.e., dorm, class, party, gym)
  • the number of people around a student through the day
  • outdoor and indoor (in campus buildings) mobility
  • stress level through the day, across the week and term
  • positive affect (how good they felt about themselves)
  • eating habits (where and when they ate)
  • app usage
  • in-situ comments on campus and national events: dimension protest, cancelled classes; Boston bombing.

We used a number of well-known pre and post mental health surveys and spring and cumulative GPA as ground truth for evaluation of mental-health and academic performance, respectively.

Below you will find papers that report on some of the findings from the StudentLife dataset. Because we are interested in spurring work in mining human behavior we have released an anonymized version of the StudentLife dataset (see below)

 

The article explains why feedback was not given to the students on the results of any stress they were feeling:

We purposely provided students with no feedback because we didn’t want to use StudentLife as a behavioral change tool. We simply wanted to “record” their time on campus. Providing feedback and intervention is the next step. For example, we might inform students of risky behavior; such as, partying too much, poor levels of sleep for peak academic performance, poor eating habits, too socially isolated, not flourishing, struggling, etc.

 

Informing students of risky behavior sounds like what mothers and fathers do every day with their children.  Students are adept in screening the human nagging out of their consciousness.  Time will tell if college students pay more attention to data from a smartphone app than they do their parents.  Whatever could go wrong with the data gathering on student movements, habits, environment, etc?  According to the article:

There are many stakeholders in student life on campus (see image above): students, faculty, student deans, docs, friends and family. All have only partial state information. We imagine StudentLife 2.0 will allow students to “connect” stakeholders by sharing their data. Such a vision represents a massive privacy problem that needs to be solved. However by connecting the stakeholders StudentLife could provide new forms of intervention to promote healthy living and safety on campus as well as help students modulate their behavior (e.g., could be as simple as not pulling all-nights and getting regular sleep) to improve GPA and life on campus.

Third parties will need personally identifiable information on students so they can ‘modulate their behavior’.  Like the data collection in K-12, students probably won’t know who has this information, where it is going and exactly for what purpose.  You can find a description of the StudentLife Data set here.  You can download the entire data set but the company cautions it is a huge file.  A brief description of the types of data contained in the data set:

The dataset directories are organized by data types. StudentLife dataset contains four types of data: sensor data, EMA data, pre and post survey responses and educational data.

Sensor Data

There are 10 subdirectories in dataset/sensing that correspond to 10 different sensor data: physical activity, audio inferences, conversation inferences, Bluetooth scan, light sensor, GPS, phone charge, phone lock, WiFi, WiFi location. All sensor data is stored in csv files.

EMA Data

You can find EMA question definitions in EMA/EMA_definition.json. Participants’ responses are stored in EMA/responses. The name of subdirectories under EMA/responses correspond to EMA question’s name. For example, EMA/responses/Stress contains all participants’ responses to the Stress EMA. Similar to sensor data, each EMA’s responses are organized by participants’ uid. You can find detailed EMA file format in EMA section

 

Pre and Post Surveys

All pre and post survey responses are stored in corresponding files under dataset/survey. The directory is organized by survey names. For example, you can find participants’ pre and post responses to PHQ-9 depression scale in survey/PHQ-9.csv. All files are in csv format, which is defined in Survey section.

Educational Data

Educational data, which include classes taken during 2013 Spring term, deadlines for each participants, grades and Piazza usage for CS65, is stored under dataset/education. Detailed description is in Educational Data section.

 

The app also records

  • when conversations occur and how long they last
  • physical activity
  • audio (silence, noise, music, unknown)
  • GPS Location
  • Bluetooth
  • Use of WIFI and location
  • Amount of ligh/darkness student is exposed to
  • Amount of time phone was locked and amount of time phone was charged
  • Seating position
  • Survey responses to questions

Here is an example of survey responses to questions:

 

(click on graphic to enlarge)

 

survey response

 

Here is information from Dartmouth’s Dr. Andrew Campbell, a computer scientist working on ubiquitous computing:

My research is focused on turning the everyday smartphone into a cognitivephone by pushing intelligence to the phone and cloud to make inferences about people’s behavior, surroundings and their life patterns.

I am interested in using smartphones to sense, inform, and nudge people in a better direction in terms of their physical and mental health; StudentLife is a good example of computational health meets smartphones at Dartmouth.

 

 

More information can be found at Forbes on this study:

For those of us who weren’t too stressed to do the reading in literature class, the sensitivity of this phone app might feel a bit Orwellian. Campbell acknowledges the potential violation of privacy. “It’s always an arms race between security and privacy,” he said. For his small experiment, he went through the school’s institutional review board and made sure the students remained anonymous.

 

But if the app will eventually be used by both students and college officials to keep mental health issues from interfering with learning, this anonymous personal information will necessarily be identifiable in the future to meet that goal.  Anybody concerned?  Here are more articles about this study:

 

Presentations

Andrew Campbell “What happens when life throws you a googly?“, Wireless Health Conference, 2014

Rui Wang “StudentLife: Assessing Mental Health, Academic Performance and Behavioral Trends of College Students using Smartphones“, ACM UbiComp, Sept 15, 2014

Rui Wang “My smartphone knows i am hungry.“, Workshop on Physical Analytics, June 16, 2014

In Press

Diagnosing depression with an app, The Independent, November 2014

Smartphone apps could be next tool for mental wellbeing, ABC (listen), Nov 2014

Mental Health App, BBC World News TV Impact, October 2014

New App Measures Students’ Moods and Mental Health, Chronicle of Higher Education, October 2014

App Can Tell When Students Are Stressed Out, September 2014

Sensing Depression, Radio interview with Nora Young on CBC/NPR Spark, September 2014

Your App, Yourself, Editorial, September 2014

Failing students saved by stress-detecting app – our work on the StudentLife study at Dartmouth is featured in the New Scientist, September 2014 – the article appeared in the printed version of NS under the headline “Phone in your feelings”

Sensitive Smart Phones Decipher The Habits Of Successful Students, September 2014

This App Passively Tracks Your Mental Health, September 2014

Stressed Out? Your Smartphone Could Know Even Before You Do, September 2014

This Phone App Knows If You’re Depressed, MIT Technology Review, September 2014

Your Smartphone Thinks You’re Sad, CBS News, September 2014

Smartphone App Knows When You’re Feeling Blue, CNET, September 2014

Dartmouth Teacher Makes Health App, Valley News, September 2014

Dartmouth’s StudentLife App Can Tell You If Your Mental Health Is Hurting Your Grades, engadget, September 2014

 

 

Published December 18, 2014

Gretchen Logue