Enhancing System Performance through Objective Feature Scoring of Multiple Persons' Breathing Using Non-Contact RF...

Area of Science:

• Biomedical Engineering
• Health Informatics
• Signal Processing

Background:

• Breathing monitoring is crucial for health sensing and disease prediction.
• Non-contact methods, particularly Radio Frequency (RF) sensing, are gaining traction for their privacy and convenience.
• Machine learning (ML) systems are increasingly used for classifying breathing abnormalities, but data dimensionality poses challenges.

Purpose of the Study:

• To develop an RF-based breathing monitoring system using software-defined radio (SDR) and channel state information (CSI).
• To classify breathing abnormalities in single and multiple-person scenarios using ML algorithms.
• To enhance system performance through optimal feature scoring.

Main Methods:

• Utilized software-defined radio (SDR) and RF sensing to capture minute variations in wireless channel state information (CSI) caused by breathing.
• Employed machine learning (ML) algorithms for the intelligent classification of breathing abnormalities.
• Applied optimal feature scoring to refine the ML models and improve performance metrics.

Main Results:

• The system successfully detected breathing abnormalities by analyzing CSI variations.
• ML algorithms achieved accurate classification in both single and multi-person scenarios.
• Optimal feature scoring led to significant improvements in accuracy, training time, and prediction speed, reaching up to 93.8% accuracy.

Conclusions:

• RF-based breathing monitoring is an effective non-contact health sensing technology.
• Optimal feature scoring is a viable solution for improving the performance of ML-based breathing abnormality classification systems.
• This technology has the potential to reduce healthcare facility stress through intelligent digital health solutions.