DEVICE IDENTIFICATION USING DEVICE FINGERPRINT AND DEEP LEARNING

Author

Prashant Baral, Southern Illinois University CarbondaleFollow

Date of Award

9-1-2021

Degree Name

Master of Science

Department

Electrical and Computer Engineering

First Advisor

Weng, Ning

Abstract

Billions of devices are connected to networks, which share data between them or with the processing device. With the advent of technology, cost-effective hardware can be interface with sensors and can act as an IoT device; This provides them with digital intelligence and communicates by connecting to Network. Like in other network devices, security is of utmost importance in IoT as well. Security in IoT is challenging than in traditional devices because of lower computing power, lower computing resources, small battery, and many IoT devices deployed. Research in cryptography, a lightweight communication protocol is focused. However, there is no substantial research effort for the identity of IoT devices which is the fundamental factor in Security. Traditional devices use MAC address, IP Address, and IMEI for physical identity but these parameters for physical device identification are prone to sniffing and spoofing; Thus, is not that reliable. Relying on these parameters for the security of IoT devices compromises the critical information and IoT device itself.The thesis utilizes the information from the packets used for communication in networks for device identification. We use IAT (Inter Arrival Time) and RTT (Round Trip Time) of packets at the router side to uniquely identify the IoT device. IAT is the time difference between two packets received at Router and RTT is the time from request to response packets at Router. IAT and RTT are different for a device as it depends on the hardware and software of a device. We plot the graph of IAT and RTT separately and use deep learning as a tool to identify a device. We use different deep learning algorithms for identification. We use CNN and LSTM for IAT graphs and RTT graphs. Both deep Learning algorithms achieve good accuracy in the classification of a device using both parameters (IAT and RTT) but to verify using the publicly available dataset, we achieved 97% accuracy in classifying using IAT using CNN in the testing dataset. Using CNN+LSTM we achieve an accuracy of 91.45% in the classification of a device using IAT as a parameter in the testing dataset.