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: default
    data_files:
      - split: train
        path:
          - phone_raw/accel/*.txt
          - phone_raw/gyro/*.txt

Smartphone and Smartwatch Activity and Biometrics 15m6

A 15.6-million-sample, multi-device time-series corpus capturing 3-axis accelerometer and gyroscope readings from 51 volunteers. Each subject performed 18 activities for three minutes apiece while wearing both a smartphone and a smartwatch, yielding synchronized sensor streams sampled at 20 Hz.

Every record is formatted as:

subject-id – unique identifier (1600–1650)
activity-code – single-letter code A–S (excluding N) denoting one of 18 activities (e.g., walking, brushing teeth)
timestamp – Unix epoch (seconds)
x, y, z – raw sensor values for the given axis

The dataset is organized into four directories (phone-acc, phone-gyro, watch-acc, watch-gyro), each containing 51 per-subject files. Raw data plus 10-second sliding-window feature vectors are provided, enabling both activity-recognition and behavioral-biometric modeling.

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