A vector convolutional deep autonomous learning classifier for detection of cyber attacks

Nowadays with the exponential rise of traffic over large scale networks, Internet is vulnerable to increased number of cyber attacks. The cyber attacks attempt to steal, alter, or destroy information through unauthorized access to systems. Recently, deep learning techniques have been proposed to detect cyber attacks. The existing deep learning based detection systems perform static detection of attacks failing to capture unknown attacks happening in evolving large network traffic. The unknown attacks could be detected on the fly if a generalizable model is designed for each evolving class of network traffic. This is effectively represented in the proposed Vector Convolutional Deep Autonomous Learning (VCDAL) classifier to detect cyber attacks in the network traffic data streams. The proposed VCDAL classifier extracts the features using vector convolutional neural network, learns the features automatically using incremental learning with distilled cross entropy, and classifies the evolving network traffic using softmax function. The proposed classifier was tested by conducting experiments on benchmark network traffic datasets and it is obvious that the proposed classifier can possibly recognize both known and unknown cyber attacks. Furthermore, it is observed from the comparative analysis that the proposed VCDAL classifier exhibits significant results compared to the existing base classifiers and state-of-the-art deep learning approaches.