---
license: mit
task_categories:
- time-series-classification
tags:
- human-activity-recognition
- sensor-data
- federated-learning
- mobile-sensing
- accelerometer
- gyroscope
size_categories:
- 1K<n<10K
---

# UCI Human Activity Recognition (HAR) Dataset

## Dataset Description

The UCI Human Activity Recognition dataset is a widely-used benchmark for human activity recognition using smartphone sensors. This dataset contains sensor readings from accelerometers and gyroscopes of smartphones worn by volunteers performing six different activities.

### Activities
The dataset includes the following 6 activities:
- **1**: WALKING
- **2**: WALKING_UPSTAIRS
- **3**: WALKING_DOWNSTAIRS
- **4**: SITTING
- **5**: STANDING
- **6**: LAYING

### Dataset Statistics
- **Number of subjects**: 30
- **Number of activities**: 6
- **Number of features**: 561
- **Training samples**: 7,352
- **Test samples**: 2,947
- **Total samples**: 10,299

## Dataset Structure

### Data Fields
- `0` to `560`: Individual sensor feature columns (561 float values total)
- `target`: Integer ID of the activity (1-6)
- `activity_label`: String label of the activity
- `subject_id`: Integer ID of the subject

### Data Splits
- **Train**: 7,352 samples
- **Test**: 2,947 samples

## Usage with Flower Datasets

This dataset is optimized for federated learning scenarios. Here's how to use it with Flower:

```python
from flwr_datasets import FederatedDataset

# Load the dataset
fds = FederatedDataset(dataset="your-username/uci-har", partitioners={"train": 30})

# Get data for a specific client (subject)
client_data = fds.load_partition(0)
```

### Federated Learning Scenarios

This dataset supports several FL scenarios:

1. **Subject-based partitioning**: Each client represents one subject (natural FL setting)
2. **Activity-based partitioning**: Clients have different activity distributions
3. **Sensor heterogeneity**: Simulate different device capabilities

## Citation

```bibtex
@misc{anguita2013public,
  title={A public domain dataset for human activity recognition using smartphones},
  author={Anguita, Davide and Ghio, Alessandro and Oneto, Luca and Parra, Xavier and Reyes-Ortiz, Jorge Luis},
  year={2013},
  publisher={European Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning}
}
```

## Original Source
- **Repository**: [UCI Machine Learning Repository](https://archive.ics.uci.edu/ml/datasets/Human+Activity+Recognition+Using+Smartphones)
- **License**: MIT

## Example Usage

```python
from datasets import load_dataset

# Load the dataset
dataset = load_dataset("your-username/uci-har")

# Access train split
train_data = dataset["train"]
print(f"Number of training samples: {len(train_data)}")

# Access a sample
sample = train_data[0]
print(f"Number of feature columns: 561")
print(f"Activity: {sample['activity_label']}")
print(f"Target: {sample['target']}")
print(f"Subject: {sample['subject_id']}")
print(f"First few features: {sample['0']:.3f}, {sample['1']:.3f}, {sample['2']:.3f}")
```

## Federated Learning Use Cases

This dataset is particularly suitable for:

- **Cross-silo FL**: Different organizations with sensor data
- **Cross-device FL**: Mobile devices performing activity recognition
- **Personalized models**: Subject-specific activity patterns
- **Non-IID scenarios**: Different subjects have different activity patterns

## Data Preprocessing

The sensor signals were preprocessed by:
- Noise filtering using median and 3rd order low pass Butterworth filters
- Sampling at 50Hz fixed-width sliding windows of 2.56 sec (128 readings/window)
- 50% overlap between windows
- Separation of gravitational and body motion components
- Jerk signals derived from body linear acceleration and angular velocity
- Fast Fourier Transform (FFT) applied to some signals
- Feature vector extraction from time and frequency domain variables