SDHAR-HOME: A Sensor Dataset for Human Activity Recognition at Home.

Area of Science:

• Gerontology
• Health Informatics
• Artificial Intelligence

Background:

• Improving the quality of life for the elderly, particularly those living alone, is a key health research objective.
• Elderly individuals may face risks at home due to physical, sensory, or cognitive limitations.
• Non-intrusive monitoring systems can help detect and mitigate potential dangers for seniors.

Purpose of the Study:

• To develop a non-intrusive home-based database for monitoring elderly residents.
• To combine sensor data, indoor positioning, and wearable activity trackers for comprehensive user monitoring.
• To validate the system's effectiveness using real-time activity recognition with advanced machine learning techniques.

Main Methods:

• Utilized a combination of non-intrusive sensors, triangulation-based indoor positioning (beacons), and activity wristbands.
• Collected two months of continuous data on the daily habits of two elderly individuals.
• Applied Deep Learning (DL) techniques, including Recurrent Neural Networks (RNN), Long Short-Term Memory (LSTM), and Gated Recurrent Unit (GRU) networks for activity recognition.

Main Results:

• Successfully labelled 18 distinct daily activities.
• Developed personalized prediction models for each user, achieving high hit rates between 88.29% and 90.91%.
• Implemented a data-sharing algorithm to enhance model generalizability and prevent neural network overtraining.

Conclusions:

• The developed system effectively monitors elderly individuals' daily activities in their homes using non-intrusive technology.
• Deep learning models demonstrate significant accuracy in recognizing activities, supporting independent living.
• The data-sharing algorithm contributes to more robust and generalizable activity recognition models for elder care.