Deep learning-Based Real-time malicious network traffic detection system for Cyber-Physical Systems
Pranjal Mestry, Ameya Rathi · 2022
Characteristics of IoT, such as large quantities and simple functions, have made it easy for IoT devices or servers to be attacked and converted into DDoS(Distributed Denial of service) attacks. The purpose of this paper is to provide an overview of security in the IoT sector and to discuss the classification of different attacks. In this paper, we have established a way to better analyze DDoS attack traffic from selected IoT devices under a local network route in real time using a hybrid deep learning model consisting of a convolutional neural network and a long term short memory neural network. Specifically, we used the CICFlowMeter tool for collecting features from real data traffic which we intercepted on the node and reduced the feature vector using feature selection algorithms. The feature selection algorithms that we used were random forest and p-value selection algorithms. The features extracted from feature selection algorithms were tested on the deep learning model and the features that gave the best performance of the model were selected. We then integrate the CICFlowMeter tool with our designed model to extract the features from real traffic and classify the nature of the traffic into normal or malicious. The results show that applying random forest feature selection algorithm leads to increase in accuracy and decrease in latency when classifying the nature of network traffic.