Related Experiment Video
Updated: Oct 24, 2025
06:37
Author Spotlight: Addressing Technical and Subjective Challenges in Measuring Classroom Attention
Published on: December 15, 2023
4.5K
Human activity recognition using wearable sensors, discriminant analysis, and long short-term memory-based neural
Md Zia Uddin1, Ahmet Soylu2,3
1SINTEF Digital, Oslo, Norway. zia.uddin@sintef.no.
Scientific Reports
|August 13, 2021
Summary
This study introduces a body sensor-based activity recognition system using deep Neural Structured Learning (NSL) and Long Short-Term Memory (LSTM). The system achieves high accuracy in recognizing daily activities, enhancing smart healthcare applications.
Area of Science:
- Biomedical Engineering
- Artificial Intelligence
- Machine Learning
Background:
- Body sensor data is crucial for smart healthcare systems, particularly for monitoring elderly individuals and detecting health risks.
- Wearable sensors offer reliable data for assessing an individual's functionality and lifestyle.
- Activity recognition systems are vital for improving eldercare and promoting independent living.
Purpose of the Study:
- To propose a novel body sensor-based activity modeling and recognition system.
- To leverage time-sequential information and deep Neural Structured Learning (NSL) for enhanced activity recognition.
- To compare the proposed system's performance against conventional machine learning methods.
Main Methods:
- Data collected from multiple wearable sensors during daily activities.
- Statistical feature processing and Kernel-based Discriminant Analysis (KDA) for feature clustering.
- Activity modeling using Neural Structured Learning (NSL) with Long Short-Term Memory (LSTM).
- Explainable Artificial Intelligence (XAI) using Local Interpretable Model-Agnostic Explanations (LIME) for decision transparency.
Main Results:
- The proposed NSL-based system achieved a recall rate of approximately 99% on a public dataset.
- Outperformed conventional methods like Deep Belief Network (DBN), Convolutional Neural Network (CNN), and Recurrent Neural Network (RNN), which yielded a maximum recall rate of 94%.
- Demonstrated robust performance in activity modeling and recognition.
Conclusions:
- The developed system offers a highly accurate and reliable method for recognizing daily activities using body sensor data.
- The system's explainability through LIME enhances trust and interpretability in AI-driven healthcare.
- This robust activity recognition system has significant potential for adoption in various environments like homes, clinics, and offices for behavior monitoring.
