Non-Contact Breath Rate Classification Using SVM Model and mmWave Radar Sensor Data

TL;DR

This paper demonstrates a non-contact method for classifying breath rates using FMCW radar data and SVM models, achieving high accuracy in distinguishing normal from abnormal breathing patterns.

Contribution

It introduces a novel non-contact breath classification system combining FMCW radar with SVM, optimizing kernel choice for improved accuracy and efficiency.

Findings

Achieved 95% classification accuracy.

Quadratic kernel required the fewest support vectors.

System effectively differentiates normal and abnormal breath rates.

Abstract

This work presents the use of frequency modulated continuous wave (FMCW) radar technology combined with a machine learning model to differentiate between normal and abnormal breath rates. The proposed system non-contactly collects data using FMCW radar, which depends on breath rates. Various support vector machine kernels are used to classify the observed data into normal and abnormal states. Prolonged experiments show good accuracy in breath rate classification, confirming the model's efficacy. The best accuracy is 95 percent with the smallest number of support vectors in the case of the quadratic polynomial kernel.