@phdthesis {2214,
	title = {Desenvolvimento dum modelo de previs{\~a}o de acidentes},
	year = {2015},
	month = {2015-01-26 00:00:00},
	school = {Universidade do Minho},
	abstract = {
Traffic crashes have placed a huge social and economic burden on society. The occurrence of crashes on rural and urban roads call for strategies to support the road administration in their efforts to reduce them. Thus, identifying factors influencing crash probability has been a leading object of research for many decades and still is. Such research targets: (i) the driver; (ii) the road and its environment; (iii) the vehicle. There is strong evidence of a need to gain better knowledge of crash prevention aspects associated to different types of road and different territorial zones using models that integrate all the known significant explanatory variables for crashes. Those include factors that influence driver performance and characteristics that represent the influence of the consistency of the geometric design of the road segment. The research developed aimed to give an answer to those questions, in particular to the context of a dense national road network integrated in a road environment characterised by a heavy land occupation, either residential or commercial and industrial. In this study, the road crash prevention models were developed using Generalized Linear Models {\textendash} GLM and Generalized Estimating Equations {\textendash} GEE. They involved des-aggregation according to the roadway elements (three-leg junctions and road segments), by different segment-lengths (200 and 400 meters) and by category of crash (total crashes and collision). Additionally, a new type of des-aggregation was introduced, a temporal des-aggregation. Verification was performed using the leave one out cross-validation method which confirmed the good performance of most of the models, where the characteristics of the road environment, like the presence of pedestrian crossings on road segments or at the three-leg junctions, are shown to play a significant role in determining the crash rate. Consequently, this study makes an important contribution to improving road safety conditions on national roads, because the crash prevention models developed can be used directly by those responsible for highway infrastructure management as a means to basing their crash rate mitigation interventions.
},
	keywords = {crash prediction models, generalized estimating equations, Generalized linear Models, National roads},
	url = {http://hdl.handle.net/1822/35827},
	author = {Costa, J. O.}
}