metadata

license: ecl-2.0
task_categories:
  - text-classification
  - feature-extraction
language:
  - en
tags:
  - public
  - text
  - tabular
  - education
  - multivariate
  - time-series
pretty_name: Smartphone and Smartwatch Activity and Biometrics 15m6
size_categories:
  - 1M<n<10M
configs:
  - config_name: dataset
    data_files:
      - split: train
        path:
          - phone_csv/accel/*.csv
          - phone_csv/gyro/*.csv
          - watch_csv/accel/*.csv
          - watch_csv/gyro/*.csv

Smartphone and Smartwatch Activity and Biometrics 15m6

A 15.6-million-sample, multi-device time-series corpus that unites 3-axis accelerometer and gyroscope streams from 51 volunteers. Each participant carried a smartphone and smartwatch while performing 18 everyday activities for three minutes apiece, generating synchronized recordings sampled at 20 Hz.

Every record is formatted as:

subject_id – integer 1600–1650 uniquely identifying the volunteer
activity_code – single ASCII letter A–S (the letter N is deliberately skipped) denoting the performed activity:
- "A" - Walking
- "B" - Jogging
- "C" - Stairs
- "D" - Sitting
- "E" - Standing
- "F" - Typing
- "G" - Brushing Teet
- "H" - Eating Soup
- "I" - Eating Chips
- "J" - Eating Pasta
- "K" - Driking from Cup
- "L" - Eating Sandwich
- "M" - Kicking (Soccer Ball)
- "O" - Playing Catch w/ Tennis Ball
- "P" - Dribbling (Basketball)
- "Q" - Writing
- "R" - Clapping
- "S" - Folding Clothes
timestamp – Unix epoch in seconds
x, y, z – instantaneous sensor readings for the corresponding device and modality

The dataset is split into four directories—phone-accelerometer, phone-gyroscope, watch-accelerometer, watch-gyroscope—each holding 51 per-subject files. Alongside raw traces, pre-computed 10-second sliding-window features are provided, enabling both activity-recognition and behavioral-biometric research.

Acknowledgements: The dataset is hosted at the UCI MLR—Smartphone and Smartwatch Activity and Biometrics