Deep Graph Convolution Neural Network based Intrusion Detection System towards Early Detection of Malicious Attacks

With the development of artificial intelligence technologies, a new network attacks and its impacts have appeared to wireless communication system, especially in high configured cloud environments. In Particular, Botnets are attack vectors through which intruders can obtain control of many systems and carry out intruding activities. Moreover, many traditional rule-based detection systems and flow based detection system have numerous solutions to identify and classify botnet attacks in real time which is circumvented by attackers. However, these mechanism have complications in keeping pace with the continuous changes of botnets on gathering the network trajectory information. In order gather the characteristics continuous behavioral changes of the botnet with intruding attacks on the set of bots against gathering the network trajectory, a novel Deep Graph Convolution Neural Network has been proposed in this paper. The current intrusion detection approach is based on deep learning architecture which has significance in identifying botnet malicious attacks effectively. Especially Mirai and Bashlite are popular botnet attacks which characterised similar to Distributed Denial of Service Attack have been captured effectively by proposed model. The proposed model is trained and evaluated on a CTU-13 malware dataset using optimization of multiple layer of the architecture using hyper parameters on max layer, hidden layers and convolution layer on features of the traffic data of the bot. Results demonstrate that the Deep Graph Convolution Neural Network can accurately and efficiently determine and classifies botnets. Experimental analysis proves proposed method is effective and efficient against conventional approaches in terms of precision and recall measures. In addition it proves that proposed model achieves high performance of 99.1 with activation function.