BandX-Activity: Human Activity Recognition Dataset with Demographics Using the MPU6050 Sensor

BandX-Activity: Human Activity Recognition Dataset with Demographics Using the MPU6050 Sensor

The BandX-Activity dataset is a comprehensive resource for human activity recognition (HAR), collected from 33 volunteers wearing the BandX wristband. This dataset includes demographic details such as age, gender, height, and weight, enabling research on personalized HAR and demography-based evaluations. Volunteers performed seven common activities: Walking (Wa), Jogging (J), Typing (T), Writing (Wr), Upstairs movement (U), Downstairs movement (D), and Cycling (C). Data was captured using the MPU-6050 sensor module, collecting accelerometer (ax, ay, az) and gyroscope (gx, gy, gz) readings at a 20 Hz sampling rate.

Dataset Contents:

Raw Data:

Each volunteer’s raw data is stored in separate files (user.csv) under the raw_data folder.

File structure:time, label, ax, ay, az, gx, gy, gz, where:

time: UNIX timestamp (IST).

label: Shortform of the activity (e.g., Wa for Walking). Full activity descriptions are stored in action_details.csv.

Accelerometer (ax, ay, az) and gyroscope (gx, gy, gz) readings.

Processed Dataset:

Processed using a 2-second window with 50% overlap.

Training Set: 15,640 samples, each with a feature shape of 40x6.

Test Set: 3,911 samples, each with a feature shape of 40x6.

Processed data is stored as processed_dataset.npz.

Includes code (sample_code/Data_Creation.ipynb) for custom processing with adjustable window size and overlap.

Metadata:

user_details.csv: Contains demographic details (age, gender, height, and weight) of each volunteer.

action_details.csv: Maps activity labels to full activity descriptions.

Code Resources:

sample_code/Data_Creation.ipynb: Preprocessing raw data into a structured format, customizable by window size and overlap.

sample_code/train_model.ipynb: Trains a 1D CNN model for HAR using the processed dataset and evaluates performance with accuracy, precision, recall, and F1-score.

Unique Features:

Personalized Data Structure: Each user’s data is stored separately, enabling research on personalized HAR.

Demographic Attributes: Includes age, gender, height, and weight, enhancing evaluations and allowing demographic-specific studies.

Comprehensive Activities: Accelerometer and gyroscope data for seven distinct activities ensure broad applicability.

TinyML Context: The dataset is collected using BandX, a low-cost, low-power wearable device powered by TinyML, making it suitable for real-time activity recognition.

Applications:

Development and evaluation of personalized activity detection models.

Analysis of activity patterns across various demographics.

Research on low-power, TinyML-enabled solutions for HAR.

Testing and benchmarking real-time, resource-constrained HAR systems.

Loading the Processed Dataset:

import numpy as np

# Load processed dataset
data = np.load("processed_dataset.npz", allow_pickle=True)
x_train, y_train = data['x_train'], data['y_train']
x_test, y_test = data['x_test'], data['y_test']

# Ensure data is numerical
x_train = x_train.astype('float32')
x_test = x_test.astype('float32')

Citation:

Please cite the following works if you use this dataset:

@inproceedings{saha2023bandx,
title={BandX: An intelligent IoT-band for human activity recognition based on TinyML},
author={Saha, Bidyut and Samanta, Riya and Ghosh, Soumya and Roy, Ram Babu},
booktitle={Proceedings of the 24th International Conference on Distributed Computing and Networking},
pages={284--285},
year={2023}
}

@inproceedings{saha2023tinyml,
title={TinyML-Driven On-Device Personalized Human Activity Recognition and Auto-Deployment to Smart Bands},
author={Saha, Bidyut and Samanta, Riya and Ghosh, Soumya Kanti and Roy, Ram Babu},
booktitle={Proceedings of the Third International Conference on AI-ML Systems},
pages={1--9},
year={2023}
}

@article{saha2024personalized,
title={Personalized Human Activity Recognition: Real-time On-device Training and Inference},
author={Saha, Bidyut and Samanta, Riya and Roy, Ram Babu and Chakraborty, Chinmay and Ghosh, Soumya K},
journal={IEEE Consumer Electronics Magazine},
year={2024},
publisher={IEEE}
}

Contact:

For any queries or collaboration opportunities, please contact:

Bidyut Saha: [e-mail]

Riya Samanta: [e-mail]