An Anomaly Aware Spatio-Temporal Graph Attention Network for Integrated Forecasting and Event Detection in Urban Traffic Streams

DOI

10.2991/978-94-6239-616-6_112How to use a DOI?

Keywords

Spatio-Temporal Traffic Prediction; Anomaly Detection; Graph Neural Network; Real-Time Traffic Management; Traffic Congestion Forecasting First Section

Abstract

The city’s traffic is difficult to predict because congestion may suddenly occur due to accidents or shutdowns of roads. Current devices are either very slow, do not understand how roads connect, or do not see these unexpected problems as they occur. We need a system that can estimate both normal traffic flows and that can find abnormal events immediately at the same time. We built a new artificial intelligence model called the Anomaly-Aware Spatio Temporal Graph Attention Network (AA-STGAT). Its main job is to do two things at once: predict traffic speed for the next 15 minutes and flag potential accidents or heavy congestion. It works by understanding the complex relationships between different roads (spatial) and how traffic changes over time (temporal). The key innovation is a special training task that teaches the model to focus on both jobs simultaneously, making it better at identifying outliers. We trained and tested it using live streams of real-world data, including traffic speeds, weather, and scheduled public events. We tested our model on real data from major cities and compared it with other advanced AI systems. Our model was the most accurate. This reduced prediction error by 12.7% and improved accident detection accuracy by 15.3%. Most importantly, it makes its predictions fast enough (in less than 90 s) to be used in real-time traffic management systems. Our model was the most accurate, reducing prediction errors by 12.7% and improving anomaly detection by 15.3% over previous state-of-the-art systems. Furthermore, our enhanced AA-STGAT++ variant achieves an additional 30.1% improvement in prediction accuracy, 7.9% higher F1-score, and 40.1% faster inference time, processing city-scale networks in under 46 seconds. This makes it highly suitable for real-time traffic management.

Copyright

© 2026 The Author(s)

Open Access

Open Access This chapter is licensed under the terms of the Creative Commons Attribution-NonCommercial 4.0 International License (http://creativecommons.org/licenses/by-nc/4.0/), which permits any noncommercial use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license and indicate if changes were made.

Download article (PDF)

TY  - CONF
AU  - Kshatriya Vinaya Sree Bai
AU  - M. Thirumaran
PY  - 2026
DA  - 2026/03/31
TI  - An Anomaly Aware Spatio-Temporal Graph Attention Network for Integrated Forecasting and Event Detection in Urban Traffic Streams
BT  - Proceedings of the International Conference on Artificial Intelligence and Secure Data Analytics (ICAISDA 2025)
PB  - Atlantis Press
SP  - 1552
EP  - 1590
SN  - 1951-6851
UR  - https://doi.org/10.2991/978-94-6239-616-6_112
DO  - 10.2991/978-94-6239-616-6_112
ID  - Bai2026
ER  -