User Security-Oriented Information-Centric IoT Nodes Clustering With Graph Convolution Networks
Anastassia Gharib, Mohamed Ibnkahla
Abstract
Information-centric Internet of Things (IoT) sensor networks allow users to access data directly from the sensing layer. This is done through cluster heads (CHs), which are selected as a result of IoT nodes’ clustering. To respond to users’ data requests, CHs aggregate, encrypt, and store locally sensed data in rounds. For data encryption, security resources are allocated to sensor nodes every round. To satisfy user security needs, very often, security resources are overutilized leading to higher energy consumption and shorter network lifetime. Meanwhile, sensor nodes’ and users’ mobility may result in link failures. Therefore, efficient clustering and security resource allocation is required to ensure users’ data and security needs are satisfied while optimizing network resource utilization. Graph convolution networks (GCNs) can help to address this challenge. GCNs perform learning on graphs while considering non-Euclidean nodes’ relations and features. Using GCNs, user awareness, and IoT nodes’ features can be incorporated into the cluster-based management of mobile information-centric IoT sensor networks. Therefore, this article proposes user-aware clustering with security resource allocation (USRA) using GCNs. In USRA, the proposed clustering algorithm improves communication reliability by optimizing users’ and nodes’ coverage. Meanwhile, the proposed security resource allocation plan prevents overutilization of security resources by considering user security needs in each cluster. Compared to existing works, USRA achieves lower energy consumption on security while ensuring high user security satisfaction. This promotes a longer network lifetime. USRA further contributes to higher communication reliability and throughput with stable data delivery latency to users.