Traffic Anomaly Detection Using Random Fourier Features and Embedded Distributions
International Journal on Advanced Science, Engineering and Information Technology · 16.10.2026 · 16 (5) · 1629-1635
DOI: 10.18517/ijaseit.16.5.14567 ↗Kurz gefasst
Fachartikel in der Zeitschrift International Journal on Advanced Science, Engineering and Information Technology, online erschienen am 16.10.2026. Verfasst von Yun-Jin Jang und 2 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
Early detection of abnormal situations in urban traffic operations—such as accidents, unexpected congestion, and roadworks—is important; however, traffic time series exhibit large normal fluctuations and sparse labels, making point-wise detection prone to increased false alarms. This paper proposes a traffic anomaly detection method that treats observations within a sliding window as an empirical distribution and detects anomalies by embedding and comparing these distributions. For each node, we aggregate observations (traffic flow, lane occupancy, and average vehicle speed) into 5-minute slots. We then approximate a Gaussian kernel using Random Fourier Features (RFF) and compute a distribution embedding as the mean of RFF features within each window. We define an anomaly score as a distance-based measure of either the change between consecutive windows or the deviation from a normal reference. The detection threshold is set via quantiles of the score distribution estimated from normal periods, enabling applicability in unsupervised or weakly supervised settings. On the XTraffic dataset with incident labels, the proposed method improves PR-AUC and F1 over statistical baselines, unsupervised learning-based methods, prediction-error-based methods, and variants without distribution embedding, while moderately reducing FAR/day and detection delay. These improvements have direct implications for public safety, as reducing false alarms mitigates alarm fatigue in traffic control centers, while earlier detection supports timely intervention to prevent secondary accidents and congestion escalation in large-scale urban road networks.
Zitieren
APA
Jang, Y., Chung, K., & Chang, J. (2026). Traffic Anomaly Detection Using Random Fourier Features and Embedded Distributions. International Journal on Advanced Science, Engineering and Information Technology, 16(5), 1629-1635. https://doi.org/10.18517/ijaseit.16.5.14567
Vancouver
Jang Y, Chung K, Chang J. Traffic Anomaly Detection Using Random Fourier Features and Embedded Distributions. International Journal on Advanced Science, Engineering and Information Technology. 2026;16(5):1629-1635. doi:10.18517/ijaseit.16.5.14567
BibTeX
@article{Jang2026,
author = {Jang, Yun-Jin and Chung, Kyungyong and Chang, JeongHyeon},
title = {Traffic Anomaly Detection Using Random Fourier Features and Embedded Distributions},
journal = {International Journal on Advanced Science, Engineering and Information Technology},
year = {2026},
volume = {16},
number = {5},
pages = {1629--1635},
doi = {10.18517/ijaseit.16.5.14567}
}
EUVIMED