<![CDATA[The roles of primary road networks in maintaining regional connectivity, supporting economic activities, and enabling emergency response, especially in disaster-prone areas, are well known. Vulnerability analysis of such networks is commonly conducted following disruption-based scenarios, in which external hazards are modeled to disrupt and degrade network performance. This study adopts a different perspective by considering vulnerability as a property that can also arise from the network’s own topological structure. In this approach, vulnerability is assumed to exist even prior to external disturbances, and thus is an intrinsic characteristic of the network. To explore this idea, a Graph Neural Network (GNN)-based framework is proposed to learn structural vulnerability directly from network topology. The primary road network of Central Java Province, Indonesia, was used as a case study. The network was modeled as an undirected, weighted graph representing two configurations: without toll roads and with toll road integration. Vulnerability labels at the node level were derived from a series of node removal experiments, capturing efficiency loss, connectivity degradation, and network fragmentation. A Graph Convolutional Network (GCN) model with two GCN layers was then trained to learn patterns associated with structural vulnerability based on node features and graph connectivity. The results indicate that structural vulnerability can be learned as a network property, with the model achieving an accuracy of approximately 82% across both configurations. This performance exceeds that of conventional centrality-based approximations. The findings also show that the inclusion of toll roads reshapes the distribution of vulnerability across the network. This supports the interpretation that toll roads act as structural modifiers by reconfiguring systemic vulnerability rather than simply increasing redundancy.]]>
Copyrights © 2026
