Textile Sensors for Measuring and Classifying Human Breathing: A Preliminary Study Using Machine Learning

To date, the assessment of breathing patterns has been carried out using different types of wearable sensors. However, there has been a limited exploration of the use of machine learning algorithms and textile sensors to develop a measurement system able to classify different breathing patterns. In this study, a wearable sensor for respiratory rate analysis was developed using a piezoresistive knitted fabric made of silver and conductive silicone. Five healthy volunteers wore two knitted fabric sensors, one at chest height and one on the abdomen. The volunteers were asked to perform four breathing patterns identified as normal breathing, apnoea, bradypnoea and tachypnoea. The change in resistance of the two sensors was measured and normalised to the normal breathing state. Features were extracted using descriptive statistics and frequency analysis. Three different structures of KNN, SVM and DNN algorithms were applied. Results showed that Linear SVM had the lowest accuracy (87.3 %) in the classification of the four breathing states. Highest accuracy (>98.0 %) was achieved by Fine KNN, Fine Gaussian SVM and Wide DNN algorithms. The results of this study are promising and could be useful for developing a wearable non-invasive textile sensor for classification of breathing patterns.