GNN-based Criticality Analysis in Interconnected Infrastructure Networks

This paper proposes a learning-based approach to identify vulnerabilities in interconnected infrastructure networks. For the first time, in contrast to purely network-based metrics (like degree and eigenvector centrality), our approach incorporates performance-based feature metrics in vulnerability assessment. Our primary goal is the creation of a scalable, adaptable framework for vulnerability analysis in interdependent critical infrastructure systems. Key to our approach is an exten-sive exploration of various metrics and their repercussions on infrastructure functionality. Secondly, the integration of feature-based metrics in a graph neural network enables us to precisely pinpoint critical nodes and links while overcoming the limitations of simulation-based methods using conventional metrics. Our framework's efficacy is validated through comprehensive experiments on the CLARC (Customizable Artificial Community) dataset. Notably, our experimental results demonstrate impres-sive accuracy rates for node classification (92.34% to 97.24%) and link classification (98.64% to 99.01%) across a diverse range of criticality metrics.