Seeing the Unseen: Learning Basis Confounder Representations for Robust Traffic Prediction

2023-11-21

AI Summary

Existing traffic forecasting models struggle with unstable X→Y mappings induced by dynamic external confounders—such as weather, accidents, and holidays—that violate stationarity assumptions. To address this, we propose STEVE, a novel framework featuring (i) a basis-vector-based continuous confounder representation mechanism, which constructs a generalizable confounder basis bank enabling zero-shot adaptation to unseen confounders via linear combination; and (ii) a confounder-agnostic relational disentanglement module that integrates graph neural networks with spatiotemporal self-supervised learning to explicitly separate confounding effects from intrinsic spatiotemporal dependencies. Evaluated on four large-scale real-world traffic datasets, STEVE consistently outperforms state-of-the-art methods and demonstrates strong robustness against both spatiotemporal distribution shifts and previously unobserved confounders. The source code is publicly available.

Abstract

Traffic prediction is essential for intelligent transportation systems and urban computing. It aims to establish a relationship between historical traffic data X and future traffic states Y by employing various statistical or deep learning methods. However, the relations of X ->Y are often influenced by external confounders that simultaneously affect both X and Y , such as weather, accidents, and holidays. Existing deep-learning traffic prediction models adopt the classic front-door and back-door adjustments to address the confounder issue. However, these methods have limitations in addressing continuous or undefined confounders, as they depend on predefined discrete values that are often impractical in complex, real-world scenarios. To overcome this challenge, we propose the Spatial-Temporal sElf-superVised confoundEr learning (STEVE) model. This model introduces a basis vector approach, creating a base confounder bank to represent any confounder as a linear combination of a group of basis vectors. It also incorporates self-supervised auxiliary tasks to enhance the expressive power of the base confounder bank. Afterward, a confounder-irrelevant relation decoupling module is adopted to separate the confounder effects from direct X ->Y relations. Extensive experiments across four large-scale datasets validate our model's superior performance in handling spatial and temporal distribution shifts and underscore its adaptability to unseen confounders. Our model implementation is available at https://github.com/bigscity/STEVE_CODE.

Problem

• Traffic Prediction
• Variable External Factors
• Model Accuracy

Innovation

• STEVE model
• self-training tasks
• disturbance factor representation