Route-Aware Dynamic Traffic Graphs for ETA: Dataset Generation and Interactive Model Testing
Guy Tordjman and Nadav Voloch
A toolchain that turns SUMO simulations and real trajectories into the same route-aware dynamic graph representation, plus a web tester for ETA models.
Machine learning for ETA needs data that aligns road topology, vehicle motion, and labels in a form graph models can learn from. We present a toolchain with two complementary paths.
Simulation-based datasets start from a SUMO network and configuration and materialize spatio-temporal graph snapshots with detailed node and edge features. Trajectory-based datasets import real traces, map-match them, and convert them into the same graph representation; Porto taxi trajectories illustrate the real-world path.
A web testing environment couples the simulator to interactive trips, model inference, and logged prediction error so ETA models can be evaluated under controlled, repeatable conditions.