Investigating Macro-Level Hotzone Identification And Variable Importance Using Big Data: A Random Forest Models Approach

Keywords

Big data; Connected Vehicle; Hotzone identification; Random forest; Variable importance; Wilcoxon test

Abstract

As Connected Vehicle technologies begin to be deployed along roadway networks, they will be providing massive amount of data. This big data can be useful in identifying safety hazardous zones, which can be complicated and unreliable today. Without sufficient data, past studies had to focus mostly on the micro-level networks. Research on macro-level hotzone identification is limited, and until this point, the contribution of various macroscopic features on the macro-level crash risks is still in dispute. This paper, with the help of massive amount of data, investigates the feasibility of using random forest for hotzone identification at macro-level - the Traffic Analysis Zone (TAZ) level. At the same time, the most influential macro-level crash risk determinants were identified by applying a series of random forest models in combination with the cross validation methods. The differences of all features between hotzones and normal TAZs were also recognized through Wilcoxon tests. Crash data of three counties in Florida during 2008 and 2009 were employed. Crash risks by different injury levels and collision types were investigated separately. Finally, the significance of various macroscopic variables was determined by different types of crash risks using variable importance analysis.The research results suggest that the distribution of road network and socio-economics are the two most important factors when proactively alleviating traffic safety issues. For developed urban areas, it is desirable to formulate specific traffic safety management strategies that accounts for zone-level socio-economics and development of road infrastructure. For zones with a higher percentage of school enrollment, pedestrian and bicycle friendly roadway system design are most beneficial. It is also desirable to take efficient countermeasures such as law enforcement and driving school training to regulate young drivers[U+05F3] behavior in school zones. For areas with high minority residence, there might be a need to use awareness campaigns in multiple languages to relieve pedestrian safety issues. Finally, additional attention should be paid to improve intersection design and management during the planning and operation processes.

Publication Date

3-12-2016

Publication Title

Neurocomputing

Volume

181

Number of Pages

53-63

Document Type

Article

Personal Identifier

scopus

DOI Link

https://doi.org/10.1016/j.neucom.2015.08.097

Socpus ID

84958936288 (Scopus)

Source API URL

https://api.elsevier.com/content/abstract/scopus_id/84958936288

This document is currently not available here.

Share

COinS