Citation

Mamdoohi S, Miller-Hooks E. J. Big Data Anal. Transp. 2022; 4(2): 153-170.

Copyright

(Copyright © 2022, Holtzbrinck Springer Nature Publishing Group)

DOI

10.1007/s42421-022-00060-9

PMID

unavailable

Abstract

Nonrecurring traffic events, including improvement actions and vehicular accidents, cause traffic congestion and travel delay. The impact of a traffic event, even when the event is quickly cleared, may last considerably longer than the event itself and may be incurred over long distances. To assess traffic impacts, identification of this larger impact area defined over time and space is required. This paper presents a statistical learning-based traffic event impact area identification method based on concepts of k-means clustering and LOcal regrESSion (LOESS). The proposed method was applied to, and evaluated on, traffic data collected by probe vehicles for three corridors within the Commonwealth of Virginia at 1-min increments over a seven-month period, producing over 14 million data records. Additional traffic incident and work zone data were also used in the analyses. The methodology's performance was evaluated through the concept of unit delay and was compared against other methods for which data requirements could be fulfilled.

RESULTS of the evaluation show that the proposed methodology accurately identifies the traffic event impact area as compared with static methods commonly used. Comparing with dynamic methods, the proposed method has simple data requirements and needs less computational effort. It is also applicable across event types with no need for calibration.

Language: en

Keywords

Data processing; Freeway work zone; k-Means clustering; Machine learning; Traffic accidents; Traffic safety