MltAuxTSPP: a unified benchmark for deep learning-based traffic state prediction with multi-source auxiliary data

Abstract

Deep learning has empowered traffic prediction models to integrate diverse auxiliary data sources, such as weather and temporal features, for enhanced forecasting accuracy. However, existing approaches often suffer from limited generality and scalability, and the field lacks a unified benchmark for fair model comparison. This absence hinders consistent performance evaluation, slows the development of robust and adaptable models, and makes it challenging to quantify the incremental benefits of different auxiliary data sources. To address these issues, we present MltAuxTSPP, a unified benchmark framework for deep learning-based traffic state prediction with multi-source auxiliary data. The framework features a standardized data container and a fusion embedding module, enabling consistent utilization of heterogeneous data and improving scalability. It produces unified hidden representations that can be seamlessly adopted by various downstream models, ensuring fair and reproducible comparisons under identical conditions. Extensive experiments on real-world datasets demonstrate that MltAuxTSPP effectively leverages weather and temporal features to improve long-term forecast performance and offers a practical and reproducible foundation for advancing research in traffic state prediction.

1. Introduction

Accurate traffic state prediction is a cornerstone of modern intelligent transportation systems (ITSs) and is essential for applications ranging from traffic management to autonomous driving. Although deep learning models have excelled at capturing complex spatiotemporal dependencies from traffic data (Do et al., 2019; Zhu et al., 2019; Lee et al., 2021), their performance is often constrained by relying solely on traffic-related data, such as the fixed position sensor data, trajectory data, network infrastructure data, and trip records data (Nagy and Simon, 2018; Shaygan et al., 2022).

Real-world traffic is significantly influenced by external factors, such as predictable temporal patterns (e.g., weekday rush hours vs. weekends) and adverse weather conditions (Essien et al., 2018; Taghipour et al., 2020). Consequently, recent studies have begun to integrate such auxiliary data to enhance prediction accuracy (Taghipour et al., 2020; Essien et al., 2021). However, the integration of multi-source data remains fraught with challenges.

Current approaches often handle a limited set of auxiliary variables, which typically lack spatial dimensions. Furthermore, these methods generally lack the generality and scalability required to accommodate diverse and growing data sources. Crucially, the field is hampered by the absence of a unified benchmark platform for systematically evaluating and comparing models under multi-source data conditions. These limitations collectively pose significant obstacles to advancing the field. The lack of generality hinders the application of existing methods to new datasets or information streams. Poor scalability makes it challenging to handle the increasing volume and variety of data. Crucially, the absence of a unified benchmark leads to fragmented research, impedes objective comparison of models, and ultimately slows ITS innovation. To overcome these obstacles, we introduce MltAuxTSPP, a novel unified benchmark framework for deep learning-based traffic state prediction using multi-source auxiliary data. At its core, MltAuxTSPP features a unified data container (UDC) and a versatile fusion embedding module (Fig. 1). This design addresses critical challenges in data handling and scalability, enabling seamless integration with various existing spatiotemporal models. This allows researchers to fairly evaluate and compare different model architectures on a standardized basis.