Hierarchical Graph Structures for Congestion and ETA Prediction

TL;DR

This paper introduces a hierarchical graph neural network approach for traffic congestion and ETA prediction, leveraging road graph topology and multi-task learning to enhance information flow and prediction accuracy.

Contribution

It presents a novel hierarchical graph neural network architecture that improves traffic prediction by incorporating graph hierarchy and multi-task learning.

Findings

Hierarchical graph structures enhance prediction accuracy.

Multi-task learning improves congestion and ETA predictions.

Code and models are publicly available for reproducibility.

Abstract

Traffic4cast is an annual competition to predict spatio temporal traffic based on real world data. We propose an approach using Graph Neural Networks that directly works on the road graph topology which was extracted from OpenStreetMap data. Our architecture can incorporate a hierarchical graph representation to improve the information flow between key intersections of the graph and the shortest paths connecting them. Furthermore, we investigate how the road graph can be compacted to ease the flow of information and make use of a multi-task approach to predict congestion classes and ETA simultaneously. Our code and models are released here: https://github.com/floriangroetschla/NeurIPS2022-traffic4cast