Machine Learning-Based Physical Activity Tracking with a view to Frailty Analysis.

Area of Science:

• Gerontology
• Biomedical Engineering
• Machine Learning

Background:

• Frailty in older adults is a significant health concern, characterized by increased susceptibility to adverse outcomes.
• It is closely linked to sarcopenia, reduced mobility, and decreased physical activity levels.
• Accurate recognition and monitoring of physical activities are crucial for understanding and addressing frailty.

Purpose of the Study:

• To investigate optimal wearable sensor types and placements for recognizing physical activities associated with frailty.
• To propose and compare two machine learning approaches for analyzing sensor data.
• To contribute to the development of effective frailty identification and prevention strategies, aligning with global health priorities.

Main Methods:

• A public dataset of physical activity signals was utilized.
• Two machine learning pipelines were developed: a deep learning approach using convolutional neural networks on activity images and a shallow learning approach with manual feature extraction and selection.
• Feature extraction involved constructing activity images from raw sensor signals for deep learning and extracting hundreds of handcrafted features for shallow learning.

Main Results:

• The study compared the performance of deep learning and shallow learning models in classifying physical activities.
• The effectiveness of different wearable sensor configurations was evaluated.
• The findings provide insights into the most discriminative features for frailty-related activity recognition.

Conclusions:

• The research offers a comparative analysis of machine learning techniques for sensor-based frailty assessment.
• Optimal sensor selection and placement are critical for accurate physical activity monitoring.
• This work is a foundational step towards developing sensor-based tools for identifying and preventing frailty, a key global health challenge.