Graph Attention Network Towards Detection and Mitigation of Distributed Denial of Service Attack to IoT Servers

Internet of Things is a revolutionizing as next generation technology which offers real time objects or things to be connected to the internet to provide effective communication among them. Despite of multiple advantages of the Internet of things, particular environment undergoes several threats like data breaches and lack of the resource availability due to its heterogeneous nature. Especially denial of service attack which leads to lack of the resource availability is serious concern in the IoT networks and servers. Botnet has been used to launch the DDOS attacks. Hence it becomes mandatory to detect and prevent the DDOS attack propagation to IoT server through IoT Network. Traditionally many intrusion detection systems have been developed to the DDOS attacks using machine learning and deep learning architectures. Those architecture does not performs well for the diversity and evolving patterns of the attacks. In this paper, a graph attention network is designed to detect and mitigate the evolving attacks patterns of the DDOS attacks through Botnet on measuring the state changes of the network traffic. Initially traffic data composed of the several attributes for normal traffic and attack traffic are preprocessed using missing value imputation and normalization approaches. Preprocessed traffic data is transformed to graph structure data. Those graph structured data is employed to graph attention network which composed of attention mechanism and convolution neural network for processing the graph transformed data. In Graph transformed data, attribute of the network traffic data is considered as node and its association between the attributes of neighboring node is considered as edge. Especially graph attention mechanism assigns different weights to the attributes represented as nodes and its association represented as edges on processing graph data through adjacency matrix. Further it efficiently captures both local and global association among the traffic information of the IoT network and IoT server on processing its attributes. Furthermore those local and global association data is projected to the activation function to linearize and aggregate it in form of attention map as it represents network patterns along its characteristics of the network data attribute. Finally attention map is classified into IoT Network attack and IoT server Attack using softmax function. Softmax function uses naïve bayes classifier to perform attack classification. Those classified attack attributes uses weight matrix to the attention coefficient to determine the volume of attack. On increased volume of attack, specified nodes were blocked as mitigation strategies. Experimental analysis and performance analysis of the graph attention network towards DDOS attack detection and attack mitigation is carried out along the conventional approaches using BoT-IoT dataset on python environment. On analysis, it is proved that current model achieves increased detection accuracy of 98.8% on identifying the DDOS attack in IOT Network and 99.1.% on identifying the DDOS attack in IOT server with high robustness on compared with state of art approaches.