An application of data mining techniques to determine the effectiveness of multiple road safety countermeasures

Al-nuaimi, Atheer Naji ORCID: 0000-0002-4350-3308 (2019). An application of data mining techniques to determine the effectiveness of multiple road safety countermeasures. University of Birmingham. Ph.D.

Al-nuaimi2019PhD.pdf
Text - Accepted Version
Restricted to Repository staff only until 1 January 2025.
Available under License All rights reserved.
Download (3MB) | Request a copy

Abstract

Road safety is a major concern worldwide and requires appropriate management. This study focuses on the effectiveness of combining road safety countermeasures to reduce car occupants’ fatalities in Low and Middle Income Countries (LMIC) using the star rating system in the international Road Assessment Programme (iRAP). Such countermeasures may be used individually or as a combination of multiple countermeasures. Yet it may be difficult for practising road safety engineers to make decisions about which combinations of countermeasures are the most effective to use when addressing road safety issues, especially with the availability of many treatments. Likely, errors in this decision-making will lead to inefficient use of limited road safety funding.

This study seeks to develop an innovative method whereby road infrastructure attributes may be combined and ranked in a new way, through processes of Knowledge Discovery in Databases (KDD). The computer environment Waikato Environment for Knowledge Analysis (WEKA) facilitated the above and enabled the extraction of knowledge in terms of decision trees and associated rules.

In addition, traditional statistical models were used to combine road infrastructure attributes in iRAP. Two statistical regression models were developed in this study, the Poisson and Negative Binomial. For the Poisson model, the results showed that overdispersion did exist as the overdispersion factor for the fatality rates data is higher than (1.0). Therefore, the Poisson model was omitted. For the Negative Binomial, the results showed that the independent variables have significant effect on the fatality numbers, as their confidence level is higher than 95%. Therefore, the Negative Binomial was shown to be an appropriate statistical regression approach to correlate road traffic and geometric attributes to fatalities numbers for the database used in this study.

The knowledge discovery and data mining techniques were found to be an enhanced alternative to the statistical models, which finds the effectiveness of road safety multiple treatments by correlating the current road attributes associated with head-on and run-off crashes without the need for historical crash data. The data mining models developed in this study were verified by comparing the mined countermeasures combinations with the real ones from the iRAP model.

Using an iRAP database from India, the most effective combinations of countermeasures for each crash type and for each road safety level (star rating) were identified. For example, it was found that the most effective countermeasures combination for road sections with head-on and run-off crashes on single lane carriageways with traversable median types and gradient ≥0% to <7.5%, requires that the differential speed limits should not be presented, medium skid resistance, medium lane width (≥2.75m to <3.25m), straight or gentle curvature (radius greater than 900m), adequate quality of curve (advance warning signs of bend and roadside indicators to indicate the curvature), short roadside distance (1 to <5m), narrow paved shoulder (≥0m to <1.0m), and adequate delineation.

The above findings could lead practising road safety engineers to combine and rank road safety treatments more satisfactorily, and could also encourage countries to invest more in road safety to save more realistic estimated fatalities in the future, especially in LMICs.

Type of Work:

Thesis (Doctorates > Ph.D.)

Award Type:

Doctorates > Ph.D.

Supervisor(s):