Related Experiment Video
Updated: Dec 26, 2025
06:32
Author Spotlight: Automated Deep Brain Stimulation for Parkinson's Disease - Exploring the Possibilities and Challenges of Home Monitoring
Published on: July 14, 2023
1.7K
Radial Basis Function Neural Network with Localized Stochastic-Sensitive Autoencoder for Home-Based Activity
Wing W Y Ng1, Shichao Xu1, Ting Wang1
1Guangdong Provincial Key Lab of Computational Intelligence and Cyberspace Information, School of Computer Science and Engineering, South China University of Technology, Guangzhou 510006, China.
Sensors (Basel, Switzerland)
|March 19, 2020
Summary
This study introduces a novel method for recognizing activities in smart homes using binary sensor data. The proposed approach achieves high accuracy, enhancing health and social care services through reliable activity recognition.
Area of Science:
- Computer Science
- Artificial Intelligence
- Ubiquitous Computing
Background:
- The Internet of Things (IoT) is increasingly integrated into daily life, particularly with smart home devices.
- Home-based activity recognition is crucial for leveraging IoT data to improve health and social care services.
- Recognizing activities involves analyzing interactions between individuals and their environment via embedded sensors.
Purpose of the Study:
- To propose a novel method for home-based activity recognition using binary sensor data.
- To enhance feature extraction from binary sensor data for improved activity recognition.
- To improve the generalization capability and robustness of activity recognition models.
Main Methods:
- Utilized binary data from anonymous sensors (pressure, contact, passive infrared).
- Proposed a radial basis function neural network (RBFNN) combined with a localized stochastic-sensitive autoencoder (LiSSA).
- Employed an autoencoder (AE) to convert binary inputs to continuous, extracting deeper features and minimizing training error and stochastic sensitivity.
Main Results:
- The proposed LiSSA-RBFNN method demonstrated superior performance across four benchmark datasets (OrdonezA, OrdonezB, Ulster, vanKasterenADL).
- Achieved high accuracy rates: 98.35% on OrdonezA, 86.26% on OrdonezB, 96.31% on Ulster, and 92.31% on vanKasterenADL.
- Outperformed established methods like Support Vector Machine (SVM), Multilayer Perceptron Neural Network (MLPNN), and Random Forest.
Conclusions:
- The LiSSA-RBFNN method is highly effective for home-based activity recognition using binary sensor data.
- The approach offers enhanced accuracy and robustness, making it suitable for practical health and social care applications.
- This work contributes a significant advancement in sensor-based activity recognition within smart home environments.
