Human Position Classification Based On Channel State Information In Wi-Fi Networks

The given work is devoted to the construction of a human position classifier based on channel state information in Wi-Fi networks. During the last decade, many researchers have been developing various monitoring systems such as human activity recognition, activity counting and activity classification based on the Wi-Fi system. This approach is attractive because Wi-Fi wireless technology is used everywhere today, therefore it does not require expensive equipment, special devices and sensors, and is also convenient to use. In this study, to implement the classifier, channel state information (CSI) is used, which describes the singularities of the signal propagation from transmitter to receiver. Based on it, classifiers are built using supervised machine learning algorithms:
$k$
nearest neighbors algorithm, logistic regression method, linear support vector machine. After preliminary work with CSI data and fine tuning of the classifiers, the obtained accuracy of correct human position classification in the Wi-Fi channel was 96.1% for the
$k$
nearest neighbors algorithm, 95.1% for the logistic regression method, and 94.6% for the linear support vector machine.