| The occurrences and outcomes of traffic crashes have long been recognized as complex events involving interactions between many factors, including not only the roadway, but also the vehicle, driver, traffic characteristics, and the environment. Numerous research efforts have been conducted in order to determine the effect of the aforementioned factors on crash occurrence and outcome. However, there is still a need for better modeling and prediction techniques. This dissertation investigates the use of innovative modeling techniques including several artificial neural networks (ANN) and statistical models to study drivers' injury severity, crash location, and future crash trends in the United States.; The multi-layer perceptron (MLP), fuzzy adaptive resonance theory (ART), radial basis functions (RBF) neural networks, nested logit, bivariate probit, and ordered probit models were applied to study drivers' injury severity at roadway sections, signalized intersections, and toll plazas. The multi-layer feed-forward neural networks (MLP and RBF) showed a superior performance to the fuzzy ART neural networks and statistical models (nested logit and ordered probit). All drivers' injury severity models showed the significance of driver's age, gender, use of seat belt, point of impact, speed, and vehicle type on the driver's injury severity level when involved in a crash. A driver's violation was significant in the cases of signalized intersections and toll plazas.; The analysis was then extended to study the location of crashes that occurred in the vicinity of toll plazas on the Orlando-Orange County expressway system using the 1999 and 2000 crash reports. Among the results, vehicles equipped with electronic toll collection (ETC) devices, especially medium/heavy duty trucks, have a higher risk of experiencing a crash at the toll plaza. Also, mainline toll plazas have a higher percentage of crash occurrences upstream of the toll plaza structure.; Finally, this dissertation investigated an application of time series models to forecast future deaths in the United States that result from passenger vehicle traffic crashes. The effects of the increase in registered light truck vehicles (light-duty trucks, vans, minivans, and sports utility vehicles—SUV) in the U.S. on the in-vehicle traffic fatality trends were incorporated in the analysis. Forecasts from the model showed that the number of annual deaths in passenger vehicle crashes stabilized around the value of 28,000 deaths. It is predicted that by the year 2010, the passenger vehicle deaths in the U.S. would be 28,600 deaths (2.1% increase compared to 2000). Also, future forecasts for the year 2010 showed that the annual deaths resulting from angle, head-on, rear-end collisions will increase by 12%, 8%, and 5%, respectively over the year 2000. Finally, the analysis was then extended to study the effect of light truck vehicles on a drivers' visibility when involved in rear-end collisions. The results showed that there are sight distance and discomfort problems when a driver is in a passenger car driving behind an LTV, which increases the likelihood of rear-end crashes for this crash configuration. |
