MotionSense: Smartphone IMU Dataset for HAR and Attribute Recognition [24 Subjects, 6 Activities]

Overview

MotionSense was created by Mohammad Malekzadeh and colleagues at Imperial College London. It contains IMU data collected from Apple iPhone 6s smartphones placed in subjects' front pants pockets via the SensingKit framework (CoreMotion API on iOS). Twenty-four subjects of different genders, heights, weights, and ages performed 6 activities in 15 trials under the same environmental conditions: downstairs (dws), upstairs (ups), walking (wlk), jogging (jog), sitting (sit), and standing (std). Data was sampled at 50 Hz.

The sensor data includes: attitude (roll, pitch, yaw), gravity vector (x, y, z), user acceleration (x, y, z — gravity-removed), and rotation rate (x, y, z from gyroscope). The dataset targets two research questions simultaneously: activity recognition from inertial data, and the inadvertent leakage of sensitive personal attributes (gender, weight) from smartphone sensor signals — a mobile privacy concern.

Column Schema

Key Statistics

• Subjects: 24 (13 male, 11 female)
• Activities: 6 (downstairs, upstairs, walking, jogging, sitting, standing)
• Sensors: Accelerometer + Gyroscope via iPhone 6s CoreMotion (iOS)
• Sampling Rate: 50 Hz
• File Format: CSV (one file per activity per subject)
• Published: 2018 — GitHub (github.com/mmalekzadeh/motion-sense) and Kaggle

Use Cases

• Human activity recognition (HAR) from smartphone IMU data
• User attribute inference (gender, weight) and mobile privacy research
• Federated learning for HAR across multiple users
• Gait and posture analysis from inertial sensors in everyday settings