Hierarchical Graph Pooling is an Effective Citywide Traffic Condition Prediction Model

TL;DR

This paper explores hierarchical graph pooling techniques to improve citywide traffic prediction by reducing data redundancy and analyzing different pooling methods and graph network inputs for enhanced accuracy.

Contribution

It demonstrates the effectiveness of hierarchical graph pooling in traffic prediction and compares two main pooling methods and input definitions for the first time.

Findings

Hierarchical graph pooling improves traffic prediction accuracy.

Node clustering pooling outperforms node drop pooling in certain scenarios.

Optimal graph network input definitions enhance model performance.

Abstract

Accurate traffic conditions prediction provides a solid foundation for vehicle-environment coordination and traffic control tasks. Because of the complexity of road network data in spatial distribution and the diversity of deep learning methods, it becomes challenging to effectively define traffic data and adequately capture the complex spatial nonlinear features in the data. This paper applies two hierarchical graph pooling approaches to the traffic prediction task to reduce graph information redundancy. First, this paper verifies the effectiveness of hierarchical graph pooling methods in traffic prediction tasks. The hierarchical graph pooling methods are contrasted with the other baselines on predictive performance. Second, two mainstream hierarchical graph pooling methods, node clustering pooling and node drop pooling, are applied to analyze advantages and weaknesses in traffic prediction. Finally, for the mentioned graph neural networks, this paper compares the predictive effects of different graph network inputs on traffic prediction accuracy. The efficient ways of defining graph networks are analyzed and summarized.