Classical Machine Learning Versus Deep Learning for the Older Adults Free-Living Activity Classification

Area of Science:

• Gerontology
• Biomedical Engineering
• Machine Learning

Background:

• Physical activity is crucial for elderly health and well-being.
• Wearable sensors offer reliable measurement of daily activities.
• Classifying activities of daily living (ADLs) is key for monitoring older adults.

Purpose of the Study:

• To evaluate machine learning and deep learning for classifying common ADLs in older adults.
• To assess the efficacy of these methods using the ADAPT dataset.
• To explore potential for profiling elderly ADL patterns in free-living settings.

Main Methods:

• Utilized classical machine learning (Support Vector Machines with ReliefF) and deep learning (LSTM networks).
• Classified activities: walking, sitting, standing, and lying.
• Validated on the ADAPT dataset with synchronized inertial sensor and video data.

Main Results:

• Both machine learning and deep learning approaches achieved high accuracy in ADL classification.
• LSTM networks and Support Vector Machines with ReliefF performed comparably.
• An F-score of approximately 97% was achieved in profiling ADLs.

Conclusions:

• Machine learning and deep learning models show significant potential for classifying ADLs in older adults.
• These methods can accurately profile ADL patterns in free-living conditions.
• Findings support the use of wearable sensors for elderly health monitoring and intervention.