WiFi-based smart sensing applications are receiving a lot of attention, and among them, intruder detection is being actively studied as security concerns increase. Recent studies on WiFi-based intruder detection systems put channel state information (CSI) data into a machine learning (ML) model and then use the output of the model to determine whether or not an intruder is present based on predefined thresholds. However, threshold-based intruder detection systems have an inherent problem in that when strict thresholds are used to increase intruder detection rate, false alarm rate that judge known users as intruders increases, and vice versa when less strict thresholds are used to reduce false alarm rate. To tackle this challenge, we propose Wi-Sniffer that increases the intruder detection rate, which is more critical than the false alarm rate by using a strict threshold, and compensates for the increased false alarm rate by putting user association (UA) information from the WiFi access point (AP) and the inference result of the ML model together in a decision tree. The UA information can be easily retrieved from the AP without additional implementation in the mobile devices. We test Wi-Sniffer in real time, and the evaluation results show that Wi-Sniffer can achieve both high intruder detection rate and low false alarm rate with only a small number of training samples and low computational complexity.
Wi-Sniffer: Wifi-based intruder detection system using deep learning and decision tree
01.06.2023
2903604 byte
Aufsatz (Konferenz)
Elektronische Ressource
Englisch
INTRUDER DETECTION DEVICE, INTRUDER DETECTION METHOD, AND INTRUDER DETECTION PROGRAM
Europäisches Patentamt | 2024
|Fence vibrations [intruder detection]
IEEE | 2000
|Online Contents | 1997