Date of Award
Doctor of Philosophy in Transportation - (Ph.D.)
Civil and Environmental Engineering
I-Jy Steven Chien
Janice Rhoda Daniel
Taha F. Marhaba
Providing a safe and efficient transportation system is the primary goal of transportation engineering and planning. Highway crashes are among the most significant challenges to achieving this goal. They result in significant societal toll reflected in numerous fatalities, personal injuries, property damage, and traffic congestion. To that end, much attention has been given to predictive models of crash occurrence and severity. Most of these models are reactive: they use the data about crashes that have occurred in the past to identify the significant crash factors, crash hot-spots and crash-prone roadway locations, analyze and select the most effective countermeasures for reducing the number and severity of crashes. More recently, the advancements have been made in developing proactive crash risk models to assess short-term crash risks in near-real time. Such models could be applied as part of traffic management strategies to prevent and mitigate the crashes. The driver behavior is found to be the leading cause of highway crashes. Nevertheless, due to data unavailability, limited studies have explored and quantified the role of driver behavior in crashes. The Strategic Highway Research Program Naturalistic Driving Study (SHRP 2 NDS) offers an unprecedented opportunity to perform an in-depth analysis of the impacts of driver behavior on crashes events.
The research presented in this dissertation is divided into three parts, corresponding to the research objectives. The first part investigates the application of advanced data modeling methods for proactive crash risk analysis. Several proactive models for segment level crash risk and severity assessment are developed and tested, considering the proactive data available to most transportation agencies in real time at a regional network scale. The data include roadway geometry characteristics, traffic flow characteristics, and weather condition data. The analysis methods include Random-effect Bayesian Logistics Regression, Random Forest, Gradient Boosting Machine, K-Nearest Neighbor, Gaussian Naive Bayes (GNB), and Multi-layer Feedforward Deep Neural Network (MLFDNN). The random oversampling technique is applied to deal with the problem of data imbalance associated with the injury severity analysis. The model training and testing are completed using a dataset containing records of 10,155 crashes that occurred on two interstate highways in New Jersey over a period of two years. The second part of the study analyzes the potential improvement in the prediction abilities of the proposed models by adding reactive data (such as vehicle characteristics and driver characteristics) to the analysis. Commonly, the reactive data is only available (known) after the crash occurs. In the proposed research, the crash analysis is performed by classifying crashes in multiple groupings (instead of a single group), constructed based on the age of drivers and vehicles to account for the impact of reactive data on driver injury severity outcomes. The results of the second part of the study show that while the simultaneous use of reactive and proactive data can improve the prediction performance of the models, the absolute crash probability values must be further improved for operational crash risk prediction. To this end, in the third part of the study, the Naturalistic Driving Study data is used to calibrate the crash risk models, including the driver behavior risk factors. The findings show significant improvement in crash prediction accuracy with the inclusion of driver behavior risk factors, which confirms the driver behavior to be the most critical risk factor affecting the crash likelihood and the associated injury severity.
Darban Khales, Sina, "Short-term crash risk prediction considering proactive, reactive, and driver behavior factors" (2021). Dissertations. 1534.