Device Fingerprinting in Wireless Networks using Deep Learning

Abstract

Device fingerprinting involves identifying devices based on attributes provided by their configuration and usage. In this work a Deep Neural Network (DNN) architecture is designed for device fingerprinting. DNN is fed Inter Arrival Time (IAT) and Transmission Time (TT) of preprocessed wireless network traffic. The DNN consists of multiple Convolution Neural Networks (CNN), Rectified Linear Units (ReLU), and maximum pooling layers. As a final step, two fully connected layers, a softmax layer and a classification layer, are applied to classify devices. To evaluate the proposed model, two benchmark datasets, SIGCOMM-2004 and SIGCOMM-2008, were used. Using only 200 frames, it can accurately fingerprint 74 devices in SIGCOMM-2004 and 48 devices in SIGCOMM-2008 with accuracy of 97.04% and 97.70% respectively. The experimental results indicate that the proposed method is more efficient, since it requires fewer frames and produces a higher level of accuracy.