Improving Machine Learning Classification Accuracy for Breathing Abnormalities by Enhancing Dataset.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Healthcare

Background:

• The COVID-19 pandemic highlighted the need for intelligent healthcare systems, especially for monitoring patients with abnormal breathing.
• Limited resources and the challenge of real-time data collection for respiratory conditions during self-isolation pose significant hurdles.

Purpose of the Study:

• To develop a non-contact, RF-based system for detecting and classifying abnormal breathing patterns indicative of COVID-19.
• To enhance machine learning model performance by generating a large dataset of simulated breathing abnormalities.

Main Methods:

• Utilized radio frequency (RF)-based technology to capture real-time breathing data.
• Employed curve fitting techniques to generate a comprehensive dataset of simulated breathing abnormalities.
• Applied and evaluated multiple machine learning algorithms for classifying eight distinct breathing patterns.

Main Results:

• The RF-based platform achieved 97.5% accuracy in classifying real-time breathing patterns.
• Incorporating simulated breathing data increased the machine learning model's accuracy to 99.3%.
• The study demonstrated improved prediction speed and reduced training time with simulated data.

Conclusions:

• The proposed RF-based system effectively detects abnormal breathing patterns, crucial for early COVID-19 indication.
• Generating simulated data is a viable strategy to overcome real-time data collection challenges and boost ML model accuracy.
• This approach offers a scalable and reliable solution for remote patient monitoring during pandemics.