ANDWELLNESS: AN OPEN MOBILE SYSTEM FOR ACTIVITY ANDEXPERIENCE SAMPLING

John Hicks, Nithya Ramanathan, Donnie Kim, Mohamad Monibi, Joshua Selsky, Mark

Hansen, Deborah Estrin

Presented by Hien Nguyen

Outline

Introduction Related works System overview Implementation Evaluation Discussion Conclusion

Introduction

AndWellness: a personal data collection system for activity and experience sampling.

Area of usage: health and behavior monitoring Example: Cancer Survivor Study by UCLA

measures the behaviors and emotions of young breast cancer survivors. Collect daily information on: nights sleep, various emotional feedback,

behaviors remind users to take simple saliva sample to

measure various biomarkers.

Introduction

In the past: asked participants to recall events used paper diaries to log events employed automated telephone systems to

record data recently used PDA or wireless mobile

devices to log data

Introduction

The trend: moving towards a real time assessment of human behavior

Because factors can and do affect the memory of participants recalling past experiences (Recall bias): current emotional state length of time asked to recall participant sitting in a foreign environment

Alleviate the issue by having participants record events as they happen or immediately thereafter

Introduction

Challenges with experiences sampling studies: Time and resources for developing robust

data collection systems from scratch Data collection systems should allow

researchers enough control to: measure a participant’s timely adherence to

the process configure when and why a participant is

queried

Introduction

AndWellness: a personal data collection system, uses mobile phones to collect and analyze data from: active, triggered user experience samples:

survey responses passive logging of onboard environmental

sensors

Related works

Can be divided into two classes: Experience sampling studies Other software systems

Related works

Experience Sampling Studies Paper diary: very low upfront cost but costly

and labor intensive post-study analysis, can not verify adherence, low motivation to adherence Reminders can help

Automated phone system: eliminates problems with data entry (response stored as integer), time stamped responses ensuring adherence

Recently, various handheld devices: prompt triggers to remind participants, compliance checks via time stamps

Related Works

Other Software Systems, some examples: Pendragon Forms, Frontline Forms, Nokia

Data Gathering: provide tools to organize surveys and collect data, but closed source and closed standards.

JavaRosa, RapidSMS, FrontlineSMS, EpiHandy: more flexible to tune for particular uses, but are primary focused on collecting textual data.

Don’t incorporate reminders, triggers etc.

Related Works

AndWellness takes advantage of the flexibility of phones to reduce participant burden: contextual triggers: avoid bothering participants

at inopportune times survey configuration: branching to avoid asking

redundant or useless questions sampling onboard sensors: i.e, GPS to collect

continuous data w/o interrupting participants. Focus on adherence and quality of responses

from non-technical participants: different from other systems.

System overview

Three subsystems: An application to collect data on an

Android mobile device A server to configure studies and store

collected data A dashboard to display participants’

statistics and data.

System overview

Paradigm: prompting participants on their mobile

device to answer surveys at configured intervals

uploading the responses wirelessly to a central server

responses are parsed into a central database

and can be viewed by both the researchers and participants in real-time.

System overview

The end-to-end data collection system contains three main components: Campaigns: contain surveys and other

continuous data collection types or sensors. Sensors: location traces (GPS) and activity

inference (still, walking, running, biking, driving) by GPS and accelerometer and uses clustering techniques

Triggers: launching surveys based on time, location, or other contextual clues, can be configured.

System overview

Dashboard view a summary of upload statistics view participant’s current progress visualize currently uploaded data: different

visualizations a map view: find relations between time,

location, and surveys responses

System overview

Example Dashboard view:

System overview

Design: has to meet a number of requirements on Usability Power Privacy Transparency

System overview

Usability: building an unobtrusive application on the mobile device

avoids interfering with standard phone operation does not drain the phone battery too quickly does not notify or require the participant’s attention

more than necessary nor exhibit high latency during participant interaction. easy to learn and use: simple clicks conditional prompts: allow branching and reducing

number of prompts to be completed automatically upload pending data

Overall: minimize interaction and interference with the user

System overview

Power: avoid having to constantly recharge the phone continuous location and activity sensor

have been optimized to balance battery drain with accuracy

tunable to allow the researcher to adjust the balance between resolution and power drain

System overview

Privacy: personal or private information to be kept securely transported using end-to-end encryption user names are created using randomized

dictionary strings to preserve participant anonyminity

informaion can be made only accessible by authorized individuals

balance between usability and privacy: how often to make the user login, to validate and to handle multiple users sharing the same phone.

System overview

Transparency: data are uploaded in near real time use dashboard to view feedback about data

collection process and to know if the device is working correctly

Data collection process becomes clear and meaningful adds motivation to the participant to continue to collect more data.

Implementation

Server implemented in Java 1.6, Spring framework hosted in the Apache Tomcat 6.0

environment uses MySQL 5.1 database HTTP based APIs control access to upload

and download

Implementation

Server

Implementation

Application on the phone implemented using standard Android

development framework in Java programming language

after reading in the configuration, automatically generates the survey questions and response inputs

user can open the settings menu to adjust the trigger times for daily triggered surveys

securely transmits the data to the server with end to end reliability, ensuring no data is lost

Implementation

Application on the phone

Implementation

Visualizations online data visualizations have been

implemented in JavaScript (data can be viewed w/o any custom software)

the system constantly pre-aggregates any collected data into much lower resolutions, grab only the necessary resolution of data

uses gzip before transport to minimize latency

Evaluation

Phone Performance: experiment using Android based mobile device: Qualcomm

MSM 7201A 528 MHz processor, 192 MB RAM, 1150 mAh lithium ion battery.

SystemSens: records CPU utilization, network usage, and current battery percentage

Evaluation

CPU Utilization: the application was set to sample activity 1/1 sec, 1/30 secs, 1/1min, and not at all

As we increase the inference rate, CPU utilization only increases a couple percent inference is quick and efficient

Evaluation

Network and Storage: again each of the activity sampling rates were selected

Evaluation

Battery Life: run the sampling activity at the above rates and measure the rate of battery drain for each the entire battery is drained in about 7.6

hours with activity inference running unfortunately fast

main battery drains: GPS and accelerometer plan to adjust the activity inference module to better duty cycles those sensors

Discussion

Discuss various issues implementing & designing AndWellness: data quality privacy transparency

Review feedback from: In lab testing initial testing for two planned studies several focus groups.

Discussion

Data Quality: Bias stems from two main sources user returning tainted data: only allow

participant to respond to survey when it triggers and not other times and no review previous survey. AndWellness has these features and new ones can be easily added.

incomplete data: due to difficulty of the question, question being triggered at inopportune times. AndWellness includes per user configuration of triggers.

Discussion

Privacy: important to allow users to trust the data collection process User login names are randomly assigned by a word

generator data linking user name to personally identifying

information is quarantined data linking can be made accessible to only specific

people, or even destroyed making the data permanently anonymous.

But too much emphasis on privacy could frustrate participant and reduced adherence rates

AndWellness can fit either need

Discussion

Transparency: Participants want to know how data is being

collected, where the data is going, and the ability to visualize the data

boost participant‘s adherence, participant trust and engagement

Discussion

Preliminary Feedback the system of triggers and reminders is

almost mandatory for continued participation

reminders and response feedback were enough to help participant feel engaged

however, the vibration pattern and ringtones used for the notifications were too annoying

Future directions

Allowing researchers to use our system to define their own campaigns, survey prompts, and data types without programmer assistance

Ability to remotely change prompts and triggers manually from the central server and even automatically based on characteristics of returned data from participants

Conclusions

AndWellness: an end-to-end data collection system designed to monitor participants’ daily habits and behaviors collects in situ behavorial and contextual

data from participants brings together the advantages from

similar systems to meet the requirements set by researchers

Question?

Thanks for listening!