Signal Dataset and Machine Learning Codes for Exercise-Associated Motion Classification Using PVDF/PVA Piezoelectric Hydrogel Sensors

Published: 24 August 2026| Version 1 | DOI: 10.17632/j53wpz5sf5.1
Contributors:
JiWei Cui, Jiayin Gao, Yuan Zhuang, Longben Huo, Zhiheng Meng, Hui Feng, Xinyuan Zhu, Xin Jin

Description

This dataset contains the signal data and machine learning codes generated for exercise-associated motion classification using wearable PVDF/PVA piezoelectric hydrogel sensors. The dataset supports the research reported in the manuscript entitled “All-polymer sea-island structured hydrogels with enhanced mechanical and piezoelectric properties for wearable fitness monitoring”. The signal dataset consists of piezoelectric output signals collected from wearable PVDF/PVA hydrogel sensors during exercise-related motion monitoring. A total of 1,200 bench-press movement trials were collected from 12 participants, including 600 standard movements and 600 non-standard movements. The dataset was established for subject-independent motion classification using a 12-fold leave-one-subject-out (LOSO) cross-validation strategy. This repository also provides the corresponding machine learning codes, including the Long Short-Term Memory (LSTM) neural network model and conventional machine learning algorithms (RBF-SVM and Random Forest) used for performance comparison. The provided codes include data preprocessing, model training, testing procedures, and evaluation workflows. The dataset and source codes enable reproducibility of the machine learning analysis and provide resources for further research on wearable piezoelectric sensors, biomechanical signal processing, and intelligent motion monitoring.

Files

Steps to reproduce

The dataset was generated using wearable PVDF/PVA piezoelectric hydrogel sensors for exercise-associated motion monitoring. The hydrogel sensors were fabricated according to the procedures described in the associated manuscript and integrated into wearable sensing devices. For signal acquisition, piezoelectric voltage signals were collected during bench-press exercises from 12 healthy participants. Each participant performed 50 standard movements and 50 non-standard movements, resulting in a total of 1,200 labeled movement trials. The collected time-series signals were assigned corresponding labels according to movement categories. To reproduce the machine learning analysis, the provided signal dataset can be used as input after preprocessing. The dataset was processed using the provided scripts for signal preparation and feature extraction. The LSTM neural network and conventional machine learning models (RBF-SVM and Random Forest) can then be trained and evaluated using the provided codes. Model performance was evaluated using a subject-independent 12-fold leave-one-subject-out (LOSO) cross-validation strategy, where signals from one participant were used as the test set while data from the remaining participants were used for model training and validation. The provided codes and datasets allow reproduction of the motion classification results reported in the associated manuscript.

Institutions

Categories

Polymer, Sensor, Biomedical Materials

Funders

  • Shanghai Municipal Science and Technology Major Project
  • National Research Foundation, Prime Minister's Office, Singapore under its Campus for Research Excellence and Technological Enterprise (CREATE) programme (CREATE Thematic Programme in Decarbonisation)
  • Joint Research and Development Center of Fluorine Materials of Shanghai Jiao Tong University and Huayi 3F
    Grant ID: 2023XYJG0001-01-06
  • Interdisciplinary Program of Shanghai Jiao Tong University
    Grant ID: YG2022QN040, YG2023QNB14
  • Advanced Materials-National Science and Technology Major Project
    Grant ID: 2026ZD0622500

Licence