STMAE: Spatial Temporal Masked Auto-Encoder for Traffic Forecasting

Xing Wu*, Chengyou Cai, Xiaoxiao Wang*, Jianjia Wang, Junfeng Yao, Quan Qian, Jun Song*

*Corresponding author for this work

Research output: Chapter in book/report/conference proceedingConference proceedingpeer-review

Abstract

The advancement of intelligent transportation systems underscores the importance of data-driven approaches in traffic forecasting, which plays a crucial role in tasks such as traffic signal control and route guidance, among others. However, the inherent uncertainty stemming from regional traffic dynamics, coupled with intricate spatio-temporal correlations, poses formidable challenges to accurate traffic prediction. Moreover, the complexities inherent in sequence forecasting across varying scales further exacerbate the accuracy dilemma. Recognizing the need for integrating information across spatial and temporal dimensions to enhance forecasting precision, a novel solution termed Spatial Temporal Masked Autoencoder (STMAE) is introduced. The STMAE framework addresses these challenges through a two-stage learning process. In the pre-training phase, an autoencoder architecture is employed to extract spatio-temporal features from the data. In the fine-tuning phase, the pre-trained encoder of the STMAE model undergoes further refinement to specifically target traffic forecasting tasks. Extensive evaluations validate the effectiveness of the proposed STMAE model. Notably, STMAE demonstrates competitive performance, achieving 3.32 Vehs MAE for long-term (60 min) traffic forecasting while operating within a reduced computational budget.
Original languageEnglish
Title of host publicationPattern Recognition
Subtitle of host publication27th International Conference, ICPR 2024, Kolkata, India, December 1–5, 2024, Proceedings, Part V
EditorsApostolos Antonacopoulos, Subhasis Chaudhuri, Rama Chellappa, Cheng-Lin Liu, Saumik Bhattacharya, Umapada Pal
Place of PublicationCham
PublisherSpringer
Pages209-223
Number of pages15
ISBN (Electronic)9783031781698
ISBN (Print)9783031781681
DOIs
Publication statusPublished - 30 Nov 2024
Event27th International Conference on Pattern Recognition - Kolkata, India
Duration: 1 Dec 20245 Dec 2024
https://link.springer.com/book/10.1007/978-3-031-78107-0 (Conference proceedings)
https://icpr2024.org/ (Conference website)

Publication series

NameLecture Notes in Computer Science
Volume15305
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349
NameICPR: International Conference on Pattern Recognition

Conference

Conference27th International Conference on Pattern Recognition
Abbreviated title ICPR 2024
Country/TerritoryIndia
CityKolkata
Period1/12/245/12/24
Internet address

Scopus Subject Areas

  • Theoretical Computer Science
  • General Computer Science

User-Defined Keywords

  • Data-driven Methods
  • Masked Autoencoder
  • Traffic Forecasting

Fingerprint

Dive into the research topics of 'STMAE: Spatial Temporal Masked Auto-Encoder for Traffic Forecasting'. Together they form a unique fingerprint.

Cite this