Extending the I-95 Rule-Based Incident Duration System With an Automated Knowledge Transferability Model. Issue 8 (August 2022)

• Record Type: Journal Article
• Authors: Huang, Yen-Lin; Lin, Yi-Ting; Chang, Gang-Len
• Is Part Of: Transportation research record. Volume 2676:Issue 8(2022)
• Page Start: 221
• Page End: 235
• Publication Date: 2022-08
• Subjects: advanced traffic management systems -- operations -- incident management -- incident management -- policy and organization -- executive management issues -- knowledge management -- knowledge management -- knowledge transfer; Transportation -- Periodicals; Roads; Transport -- Périodiques; Routes -- Périodiques; Routes -- Conception et construction -- Périodiques; Roads; Transportation; 388.05
• DOI: 10.1177/03611981221083292
• Languages: English
• ISSNs: 0361-1981

Abstract

The rule-based incident duration prediction model (IDPM), covering Interstate highways I-95, I-495, and I-695, has been adopted by the Maryland Department of Transportation State Highway Administration in its daily responses to non-recurrent congestion. In light of its effectiveness and robustness in practice, expanding such a system to all other highways emerges as desirable but a challenging task, because of the need to integrate field operators' expertise in generating prediction rules and the dependence on sufficient incident records for key parameter calibration. To circumvent such a data-demanding and time-consuming process for knowledge acquisition and refinement for extending the IDPM's spatial coverage, this study has proposed a knowledge transferability analysis (KTA) method, featuring its automated process to assess, select, and transfer existing prediction rules to perform incident duration estimate for the new target highway. Evaluation of the proposed KTA with the incident records from Maryland I-70, using both transferred and customized local rules, reveals that it can achieve accuracy of 87% with the training dataset (i.e., 2016–2018) and 82% with the test dataset (i.e., 2019), comparable to the current system's performance but demanding much fewer incident records for model calibration and significantly less effort for system expansion.