Date of Award
Doctor of Philosophy in Transportation - (Ph.D.)
Civil and Environmental Engineering
I-Jy Steven Chien
Janice Rhoda Daniel
Taha F. Marhaba
Motor vehicle crashes are one of our nation's most serious social, economic and health issues. They are the leading cause of death among children and young adults, killing approximately 1.35 million people each year. Providing a safe and efficient transportation system is the primary goal of transportation engineering and planning. To help reduce traffic fatalities and injuries on roadways, crash prediction models are used to forecast the injury severity of potential crashes and apply precautionary countermeasures accordingly. Most of these models are reactive as they use historical crash data to categorize crash-related factors. Recently, advancements have been made in developing proactive crash prediction models to measure crash risk in real-time.
Crash occurrence and the resulting injury severity are influenced by several stochastic factors including driver behavior characteristics, roadway characteristics, vehicle characteristics, traffic volumes, environmental conditions, and time conditions. The objective of this research is to develop a data-driven model for crash injury severity prediction using the aforementioned factors intended to support highway safety improvement projects. The model interacts with various data sources in effective and efficient manners, which are expected to support state and local traffic management agencies in planning and operations to reduce crash injury severity.
This research explores several types of data and modeling techniques used in crash studies. The data associated with crashes on New Jersey freeways in 2017 are collected along with INRIX reported speeds. The weighted speed variance across the traffic stream before crash occurrence is introduced as a potential variable affecting crash injury severity in the prediction model. An Artificial Neural Network (ANN) is developed to estimate crash injury severity based on potential risk parameters suggested by previous studies and data availability for New Jersey freeways. A linear regression model (LRM) is also developed using the same dataset and the performance of both models are compared and discussed. While both models have advantages and limitations, the ANN outperforms the LRM for all levels of injury severity. In addition, the traffic speed and the weighted speed variance are two variables that highly influence the injury severity level resulting from a crash.
The model can be used both proactively and reactively. It can be integrated into the State Strategic Highway Safety Plan (SHSP) to allow highway safety programs and partners in the State to work together to align goals, leverage resources and collectively address the State's safety challenges. The ability to estimate crash injury severity in real-time allows transportation agencies to deploy active countermeasures to increase safety and reduce crashes and associated delays and costs. These countermeasures include increasing service patrol coverage, implementing stricter speed rules and lowering dynamic speed limits under critical conditions to avoid crash resulting injuries and fatalities and enhance emergency response time in case of a crash.
Abisaad, Rima, "Crash injury severity prediction with artificial neural networks" (2021). Dissertations. 1563.