Tracking vital signs of a patient using channel state information and machine learning for a smart healthcare system

In a smart healthcare system, the sensor-embedded wearable devices have the ability to track various vital signs of a patient. However, such devices need to be worn by the patients all the time. These devices have limitations such as their battery lifetime, charging mechanism, and hardware-related cost. Moreover, these devices transmit a huge amount of redundant and inconsistent data. The transmitted data need to be fused to remove any outlier so that only highly- refined data are available for decision-making. In this paper, we use channel state information (CSI) to track the vital signs of a patient and remove any outliers from the gathered data. We monitor the respiration rate of a patient during sleep with minimal hardware-related cost. Our CSI-based approach no longer requires the patients to wear any wearables and can monitor even the minute fluctuations in a WiFi signal. For extracting useful features from the respiratory data, three types of feature extraction techniques are used. In order to select important features from the extracted feature space, three feature selection algorithms, i.e., Relief, mRMR, and Lasso, have been investigated. In addition, for predicting the health conditions of a patient, four machine learning classification algorithms, i.e., KNN, SVM, DT, and RF, are utilized. The use of CSI ensures that highly refined and fused data are available for feature selection, and the selected features are presented to the ML classification algorithms for predicting the health condition of the patient.