Real-Time Stream Learning System for Monitoring Activities of Daily Living in Older Adults

Area of Science:

• Gerontology
• Computer Science
• Biomedical Engineering

Background:

• Monitoring activities of daily living (ADLs) is crucial for assessing older adults' functional status and autonomy.
• Traditional methods like questionnaires and offline machine learning models lack adaptability to dynamic environments and natural human movement variability.
• Real-time monitoring systems are needed to overcome the limitations of current assessment approaches.

Purpose of the Study:

• To design, implement, and evaluate a real-time data collection and processing system for ADL classification using Stream Learning (SL) models.
• To compare the performance of SL models with conventional offline models in real-world conditions.
• To determine if SL-based monitoring provides more reliable and responsive metrics than traditional offline approaches.

Main Methods:

• Developed a real-time system using a mobile app to capture accelerometer and gyroscope data.
• Transmitted data streams to a cloud server for segmentation, alignment, storage, and incremental model updates.
• Evaluated both offline and SL models using metrics like precision, adaptability, and prediction stability.

Main Results:

• The system enabled continuous real-time data acquisition and incremental model training.
• Offline models showed competitive accuracy but limited adaptability to variations.
• SL models demonstrated more robust, stable, and adaptable predictions to the natural variability of daily activities.

Conclusions:

• The integration of mobile sensing and SL is feasible for continuous ADL monitoring in real-world settings.
• SL approaches offer superior adaptability, stability, and responsiveness compared to traditional offline models.
• Hoeffding Tree model balanced accuracy and latency effectively, highlighting SL's potential for scalable, low-cost functional monitoring.