Dissertations and Theses @ UNI
Availability
Open Access Thesis
Keywords
Traffic accidents--Iowa; Cluster analysis; Traffic accidents; Iowa; Academic theses;
Abstract
In the US, crashes are a leading cause of death and disability. To effectively counter this problem, measures are aimed at locations with high crash risks. One common method to determine these "hazardous" locations is to identify clusters of crashes, also referred to as hotspots or blackspots. Several methodologies have been developed to achieve this aim. However, there is no consensus as to what method(s) is "correct." This study evaluates some of the methodologies in the CrimeStat statistical program that are used to cluster crashes with the aim of determining those most suitable for certain crash types (younger drivers, failure to yield right of way, and fixed object crashes) at different crash levels of analysis (point, line and polygon). Viable crash clusters generated were selected based on five criteria: visual assessment; compactness of crashes within and between identified clusters; consistency of cluster points/areas; the reproducibility of the clusters; and simulation results. The methodologies were evaluated based on: visual assessment of cluster results; reproducibility capability; simulation capacity; flexibility of methodologies; and their capacity for adjustment. The methodologies in CrimeStat were found to be effective in determining crash clusters. The fuzzy mode was evaluated as the best methodology for point clustering while nearest neighbor hierarchical clustering and kernel density estimation were both adjudged effective in linear clustering. The kernel density was selected as the most suitable for area clustering. Annual average daily traffic data was used to adjust for crash rates as an important evaluation process. Some rules of thumb to guide the use of these methodologies were also determined and recommended. Some problems faced include the program's use of Euclidean space to measure crash clusters, which are basically linear events and features, and the acceptance and use of only point shapefiles. Others include inadequacy in available data and, importantly, the subjectivity involved in defining and measuring crash clusters. Future studies should attempt to determine statistical relationships among clustering parameters. It is recommended that a better concept of space that will deal with the linearity of crash events is necessary for greater accuracy.
Year of Submission
2005
Degree Name
Master of Arts
Department
Department of Geography
First Advisor
Tim Strauss
Second Advisor
Thomas Fogarty
Third Advisor
Ramanathan Sugumaran
Date Original
2005
Object Description
1 PDF file (191 leaves)
Copyright
©2005 Leonard Sitji Bombom
Language
en
File Format
application/pdf
Recommended Citation
Bombom, Leonard Sitji, "An Evaluation of Crash Cluster Identification Methodologies: A Case Study of Iowa" (2005). Dissertations and Theses @ UNI. 2119.
https://scholarworks.uni.edu/etd/2119
Comments
If you are the rightful copyright holder of this thesis and wish to have it removed from the Open Access Collection, please submit a request to scholarworks@uni.edu and include clear identification of the work, preferably with URL.