Evaluating Multi-Sensor Placement and Neural Network Architectures for Physical Activity Level Classification.

Area of Science:

• Biomedical Engineering
• Wearable Technology
• Osteoarthritis Management

Background:

• Accurate physical activity level (PAL) monitoring is crucial for effective osteoarthritis (OA) management.
• Current methods may lack precision in distinguishing activity intensities.

Purpose of the Study:

• To evaluate the impact of sensor placement and deep learning models on PAL classification accuracy.
• To determine optimal configurations for wearable activity monitoring in OA patients.

Main Methods:

• Utilized deep learning models, including CNN-LSTM, for PAL classification.
• Compared single-sensor (wrist) versus multi-sensor (wrist and ankle) configurations.
• Assessed classification accuracy using Metabolic Equivalent of Task (MET) values.

Main Results:

• The wrist-and-ankle (WA) configuration significantly improved high-intensity activity classification from 53% to 86.2% compared to wrist-only.
• The CNN-LSTM model achieved the highest overall accuracy at 95.09%.
• Multi-sensor configurations demonstrated statistically superior performance over single-sensor setups (p < 0.05).

Conclusions:

• Multi-sensor wearable systems, particularly with wrist and ankle placement, enhance PAL classification accuracy.
• The WA configuration provides a practical balance of usability and precision for OA management.
• Deep learning models combined with optimized sensor placement offer a cost-effective approach to activity monitoring.