<![CDATA[Smart city infrastructures such as transportation networks, energy grids, and water distribution systems are increasingly equipped with heterogeneous sensors that generate large-scale, interconnected data. However, predicting infrastructure reliability remains challenging due to the complex spatial and temporal dependencies within these networks. This research proposes a Graph Neural Network (GNN)-based framework designed to model urban infrastructure as a graph consisting of nodes (e.g., intersections, substations, sensors) and edges (e.g., roads, pipelines, power lines), each enriched with multimodal operational features. By leveraging message-passing mechanisms and spatiotemporal GNN architectures, the model effectively learns relational patterns and evolving system dynamics to predict node and edge failure risks. Experimental results show that the proposed GNN significantly outperforms traditional machine-learning models, time-series approaches, and standard neural networks, achieving higher accuracy, lower error rates, and stronger generalization across infrastructure domains. Visual analyses including graph heatmaps, spatial propagation patterns, and critical node detection demonstrate the model’s ability to identify vulnerability clusters and potential cascading failures. The learned graph embeddings provide interpretable insights into system behavior, highlighting key risk factors and influential structural components. The findings suggest major real-world implications, including improved early warning systems, smarter maintenance scheduling, and substantial cost savings for urban management. While the framework’s performance depends on sensor data quality and computational resources, the study highlights the strong potential of graph-based learning to support more resilient, proactive, and data-driven smart city infrastructure management.]]>
