UCI Human Activity Recognition Using Smartphones (UCI-HAR)
UCI-HAR records 30 volunteers aged 19 to 48 doing six daily activities (walking, walking upstairs, walking downstairs, sitting, standing and lying) with a Samsung Galaxy S II worn on the waist. The phone's accelerometer and gyroscope captured 3-axis acceleration and angular velocity at 50 Hz, and labels were added manually from video. The data are split 70/30 by volunteer into training and test sets.
- Device: Samsung Galaxy S II (waist)
- Labels: 6 activities, video-annotated
- Size: 10,299 instances
- Hosted: UCI ML Repository (2012)
Original description
Human Activity Recognition database built from the recordings of 30 subjects performing activities of daily living (ADL) while carrying a waist-mounted smartphone with embedded inertial sensors.
The experiments have been carried out with a group of 30 volunteers within an age bracket of 19-48 years. Each person performed six activities (WALKING, WALKING_UPSTAIRS, WALKING_DOWNSTAIRS, SITTING, STANDING, LAYING) wearing a smartphone (Samsung Galaxy S II) on the waist. Using its embedded accelerometer and gyroscope, we captured 3-axial linear acceleration and 3-axial angular velocity at a constant rate of 50Hz. The experiments have been video-recorded to label the data manually. The obtained dataset has been randomly partitioned into two sets, where 70% of the volunteers was selected for generating the training data and 30% the test data.
The sensor signals (accelerometer and gyroscope) were pre-processed by applying noise filters and then sampled in fixed-width sliding windows of 2.56 sec and 50% overlap (128 readings/window). The sensor acceleration signal, which has gravitational and body motion components, was separated using a Butterworth low-pass filter into body acceleration and gravity. The gravitational force is assumed to have only low frequency components, therefore a filter with 0.3 Hz cutoff frequency was used. From each window, a vector of features was obtained by calculating variables from the time and frequency domain.
Check the README.txt file for further details about this dataset.
A video of the experiment including an example of the 6 recorded activities with one of the participants can be seen in the following link: http://www.youtube.com/watch?v=XOEN9W05_4A
An updated version of this dataset can be found at http://archive.ics.uci.edu/ml/datasets/Smartphone-Based+Recognition+of+Human+Activities+and+Postural+Transitions. It includes labels of postural transitions between activities and also the full raw inertial signals instead of the ones pre-processed into windows.
Verbatim fromUCI Machine Learning Repository· Licence: CC BY 4.0 (UCI default)
UCI Human Activity Recognition Using Smartphones (UCI-HAR): overview
UCI-HAR (Anguita et al., 2013) is the most widely used smartphone activity-recognition benchmark. Thirty volunteers aged 19 to 48 wore a Samsung Galaxy S II on the waist while doing six activities; the phone's accelerometer and gyroscope were recorded at 50 Hz and cut into 2.56 s windows with 50% overlap. Each window comes both as raw inertial signals and as a 561-feature vector, with an official split of 21 training and 9 test volunteers. It has no PPG, but it is a standard reference for motion models used alongside wearable PPG.
What's inside
- Subjects
- 30 volunteers, aged 19–48
- Activities
- 6: walking, walking upstairs, walking downstairs, sitting, standing, laying
- Sensors
- Smartphone (Samsung Galaxy S II) on the waist: 3-axis accelerometer and gyroscope at 50 Hz
- Windows
- 2.56 s with 50% overlap (128 samples); body/gravity split with a 0.3 Hz Butterworth filter
- Features
- 561 time- and frequency-domain features per window, plus raw inertial signals
- Instances
- 10,299 windows: 7,352 train, 2,947 test
- Split
- Official subject split: 21 volunteers train, 9 test
- Size
- 58.2 MB zip
How to access it
Host: UCI Machine Learning Repository
- Download the zip (58.2 MB) from the UCI page; no registration.
- Unzip the inner 'UCI HAR Dataset.zip'.
- Raw windows are in train/ and test/ 'Inertial Signals'; features are in X_train.txt / X_test.txt.
Licence: Creative Commons Attribution 4.0 International (CC BY 4.0)
What it allows: Sharing and adaptation for any purpose, including commercial, with credit.
Load UCI Human Activity Recognition Using Smartphones (UCI-HAR) in Python
pip install numpy pandasimport numpy as np
import pandas as pd
root = "UCI HAR Dataset"
def load(split):
# 2.56 s windows (128 samples at 50 Hz), 9 raw channels
chans = [f"{s}_{a}" for s in ("body_acc", "body_gyro", "total_acc") for a in "xyz"]
X = np.stack([np.loadtxt(f"{root}/{split}/Inertial Signals/{c}_{split}.txt") for c in chans], axis=-1)
y = np.loadtxt(f"{root}/{split}/y_{split}.txt", dtype=int) # 1..6
subject = np.loadtxt(f"{root}/{split}/subject_{split}.txt", dtype=int)
return X, y, subject
X_train, y_train, s_train = load("train")
X_test, y_test, s_test = load("test")
labels = pd.read_csv(f"{root}/activity_labels.txt", sep=" ", header=None, index_col=0)[1]
print(X_train.shape, X_test.shape, len(set(s_train)), len(set(s_test)), labels.tolist())Run on the UCI download on 16 Oct 2026: 7,352 × 128 × 9 train and 2,947 × 128 × 9 test windows, 21 and 9 subjects.
Caveats
- Windows overlap by 50%, so a random window split leaks data between train and test; use the official subject split or subject-wise CV.
- Sitting and standing are the most often confused classes.
- The 561 features were computed from the same windows, so feature-based and raw-signal results are not directly comparable.
- Postural transitions are not labelled here; the later Smartphone-Based Recognition of Human Activities and Postural Transitions dataset adds them with full raw signals.
- Phone on the waist, not the wrist: results do not transfer directly to smartwatch PPG devices.
Similar datasets
| Dataset | Subjects | Signals | Licence |
|---|---|---|---|
| WISDM AR v1.1 | 36 | accel | Not stated |
| HHAR | 9 | accel | Creative Commons Attribution 4.0 International (CC BY 4.0) |
| mhealth | 10 | accel | Creative Commons Attribution 4.0 International (CC BY 4.0) |
| PAMAP2 | 9 | accel, gyro, magnetometer, skin-temp, hr | CC BY 4.0 |
| USC-HAD | 14 | accel | Not stated |
Sources: UCI-HAR on UCI · Anguita et al. 2013 (ESANN) · Updated 2026-10-16
Key numbers
How it works
No model in the hub reports results on this dataset yet.
- Device roles
- Study wearable (IMU)
Leaderboard rows· 3
All leaderboards| Task | Dataset | Metric | Value | Rank |
|---|---|---|---|---|
| Human activity recognition (IMU) | UCI Human Activity Recognition Using Smartphones (UCI-HAR) | f1_har_unspecified ↑ | 98.55 | #1 / 3 |
| Human activity recognition (IMU) | UCI Human Activity Recognition Using Smartphones (UCI-HAR) | f1_har_unspecified ↑ | 97.53 | #2 / 3 |
| Human activity recognition (IMU) | UCI Human Activity Recognition Using Smartphones (UCI-HAR) | f1_har_unspecified ↑ | 94.61 | #3 / 3 |
Models trained or evaluated on it
FAQ
How many subjects and activities are in UCI-HAR?
30 volunteers aged 19–48 and 6 activities: walking, walking upstairs, walking downstairs, sitting, standing and laying.
What is the UCI-HAR train/test split?
The dataset ships with a subject split: 21 volunteers (7,352 windows) for training and 9 volunteers (2,947 windows) for testing.
What sampling rate and window length does UCI-HAR use?
50 Hz accelerometer and gyroscope, segmented into 2.56 s windows (128 samples) with 50% overlap.
What is the UCI-HAR licence?
CC BY 4.0 on the UCI Machine Learning Repository.
Cite
@misc{ucihar2026,
title = {{UCI Human Activity Recognition Using Smartphones (UCI-HAR)}},
year = {2026},
howpublished = {UCI ML repository},
doi = {10.24432/C54S4K},
url = {https://archive.ics.uci.edu/dataset/240/human+activity+recognition+using+smartphones}
}History
Record updated