United States Department of Transportation
i
Leveraging Probe Data for Improving Incident Management Practice in Rural Areas
-
2024-11-01
-
[PDF-2.59 MB]
Details:
-
Creators:
-
Corporate Creators:
-
Corporate Contributors:
-
Subject/TRT Terms:
-
Publication/ Report Number:
-
Resource Type:
-
Geographical Coverage:
-
Edition:Final Report (November 2023–November 2024)
-
Corporate Publisher:
-
Abstract:Traffic data are essential for decision-making by state departments of transportation in planning, designing, operating, maintaining, and rehabilitating transportation systems. However, collecting traffic counts at numerous portable sites in rural areas demands significant time and resources. In response, the Georgia Department of Transportation (GDOT) has been exploring alternative data acquisition technologies to efficiently gather traffic data across Georgia’s rural road network. With the increasing availability and use of probe data in various transportation applications, this study examines the feasibility of leveraging probe data for two key purposes: (1) improving vehicle miles traveled (VMT) reporting and (2) enhancing incident management practices in rural areas. To evaluate the feasibility of VMT reporting, traffic volumes estimated from probe data on rural state roads were compared to traffic volumes from GDOT’s portable count sites, which served as the ground truth. Using a sample of 500 portable count sites in rural South Georgia, probe-derived traffic volumes yielded an overall estimation error of 21 percent and 29 percent for daily vehicle miles traveled (DVMT) based on data from Vendor 1 and Vendor 2, respectively. Notably, the most stable traffic estimates occurred on Wednesdays; estimating DVMT using only Wednesday’s data reduced the error to −4 percent for Vendor 1 and 5 percent for Vendor 2. To enhance incident management, event data from the Regional Integrated Transportation Information System (RITIS) were employed to model both the risk and duration of incidents on interstate highways in rural South Georgia, patrolled by Georgia’s Coordinated Highway Assistance and Maintenance Program (CHAMP). Incident risk and duration were treated as binary classification problems, utilizing a state-of-the-art gradient-boosting tree method. The incident risk model achieved an F1 score of 0.65 with a recall of 0.74. For incident duration, a 30 min threshold yielded the best classification performance, with an F1 score of 0.72. Feature importance analysis, combined with spatiotemporal heatmaps, uncovered specific patterns that can inform and optimize incident management practices.
-
Format:
-
Funding:
-
Collection(s):
-
Main Document Checksum:
-
Download URL:
-
File Type:
