MyMove Dataset

Overview

Publications

If you use this dataset in your research, please cite the following papers:

@inproceedings{kim2022mymove,
  title     = {MyMove: Facilitating Older Adults to Collect In-Situ Activity Labels on a Smartwatch with Speech},
  author    = {Kim, Young-Ho and Chou, Diana and Lee, Bongshin and Danilovich, Margaret and Lazar, Amanda and Conroy, David E. and Kacorri, Hernisa and Choe, Eun Kyoung},
  booktitle = {Proceedings of the 2022 CHI Conference on Human Factors in Computing Systems},
  year      = {2022},
  doi       = {10.1145/3491102.3517457}
}

@article{khayami2025groundtruth,
  title      = {From Verbal Reports to Personalized Activity Trackers: Understanding the Challenges of Ground Truth Data Collection with Older Adults in the Wild},
  author     = {Khayami, Hossein and Wang, Lining and Kim, Young-Ho and Lee, Bongshin and Conroy, David E. and Lazar, Amanda and Choe, Eun Kyoung and Kacorri, Hernisa},
  journal    = {Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies},
  volume     = {9},
  number     = {2},
  articleno  = {38},
  year       = {2025},
  doi        = {10.1145/3731749}
}

Study Design & Data Collection Protocol

The study was an in-the-wild deployment conducted in May–July 2021. Participants collected their own activity data in their natural home and community environments in the Northeast region of the United States. Due to COVID-19, all onboarding and debriefing sessions were held remotely via Zoom. Study equipment was delivered and picked up by a researcher.

Timeline

MyMove App Interface

Participants reported activities using MyMove, a custom Android Wear app. They could initiate a report voluntarily or respond to an ESM prompt, describe their activity, duration, and effort level via speech, review the recording, and submit — all from the wrist.

Labeling Pipeline

Verbal reports and thigh-worn sensor data were triangulated to produce consensus activity labels. The figure below illustrates how data from both sources flows through the labeling pipeline.

Activity Label Taxonomy

Both activPAL and verbal reports were mapped to a hierarchical activity scheme aligned with sensor capabilities:

Activity
├── Sedentary
│   ├── Lying
│   └── Sitting
│       ├── Sitting in transport
│       └── Sitting not in transport
├── Upright
│   ├── Standing
│   └── Stepping
│       ├── Low Stepping
│       ├── Moderate Stepping
│       └── Vigorous Stepping
└── Cycling

Participants

Individual Participant Details

MVPA = moderate-to-vigorous physical activity (≥100 steps/min); Sedentary = time spent sitting and lying while awake. Measured by activPAL during the 7-day data collection period.

Inclusion Criteria

• Aged 60 or older
• Comfortable describing their activity in English
• Curious about their activity levels and interested in data collection
• No severe speech, hearing, motor, movement, or cognitive impairments
• Stable home Wi-Fi and able to join Zoom video calls
• Right-handed (required by the Fossil Gen 5 button layout)

Raw Sensor Data

Smartwatch IMU (via MyMove App)

Each record corresponds to one 60-second window. IMU signals are recorded as a 10-second burst at 50 Hz every minute.

Note: Total (raw) acceleration, as commonly used in HAR datasets, equals Linear Acceleration + Gravity and can be reconstructed by summing the two signals above.

Data format: Protocol Buffers, stored on-watch and uploaded to a MySQL database via the companion phone's Wi-Fi.

activPAL4 Thigh-Worn Sensor

Verbal Reports

Data Visualization

IMU Signal Samples

One 10-second window of all four IMU signals — linear acceleration, gravity, rotation vector, and magnetometer — recorded from the smartwatch during five representative activities: lying, sitting, standing, walking, and running.

Raw IMU sensor readings across five activities (lying, sitting, standing, walking, running).

Verbal Reports Aligned with Sensor Data

Examples of verbally reported activity segments aligned with activPAL posture classifications and smartwatch step counts over time.

Personalized Stepping Intensity

Stepping cadence distribution for one participant (P13), showing distinct peaks corresponding to low, moderate, and vigorous stepping intensities. Thresholds vary across individuals and require personalization.