Paradiso S., Ziakopoulos A., Yannis G., “A Dual Graph Framework for Edge Embedding and Clustering in Road Safety Analysis”, Proceedings of the Transport Research Arena TRA 2026 Conference, Budapest, 18-21 May 2026.

A road network can be represented as a graph, and telematics data can be spatially aggregated onto it, enabling the capture of driving behaviors across the network. Consequently, Graph Neural Networks (GNNs) provide a robust framework for analyzing such data and enabling network-level inference. While extensive research has focused on nodes, working with edges, being non-point entities, presents additional challenges. This study adopts a dual graph approach, converting edges into nodes, to perform node embedding by using a GNN model on the main central area of Athens, integrating geometric and telematics data capturing both the spatial structure of the road network and driving behavior patterns. To partition the network, K-Prototypes was first applied to raw edge attributes to handle mixed variable types, while K-Means was applied to the numerical embeddings generated by the GNN. Clustering GNN-generated embeddings with K-Means outperformed K-Prototypes, showing stronger separation and more reliable partitioning of road networks, which can support proactive safety planning and infrastructure management.