Recognizing Node Intrusion Tendencies in IoT Environments via Deep Learning and Network-Level Feature Analysis

Abstract

The IoT environment is becoming more and more susceptible to intrusion attacks due to its decentralized infrastructure, fewer security boundaries and heterogeneity and dynamism of the network. Discovering malicious parties in these environments is a challenging task, mainly due to the mobility of the nodes and the intermittent contact between them. This study introduces a deep learning-based approach to detect intrusion tendencies on the node level based on the analysis of essential network-level features. The link duration, self-healing latency and number of packets that potential attacker nodes receive were obtained from an analytic network profiling model. These were then employed in the training and testing of the three deep learning models: Feedforward Neural Network (FFNN), Cascade Backpropagation Neural Network (CBPNN), and Convolutional Neural Network (CNN). Of those, the CNN showed the best performance, obtaining intrusion detection accuracy of 85.5%. The proposed approach emphasizes the importance of combining network behavior analytics with deep learning technologies to increase the security of IoT environments.