ESTIMATION OF POINT PROCESSES ON DIRECTED GRAPHS

Abstract: Road accidents are one of the most important concerns worldwide. Studying and analysing them is one way to improve and ensure safety on the roads. Traffic accident records for four years have been provided and are analysed in this thesis. The dataprocessing is done according to the metadata and data definitions of the European framework for road accident. On top of this, a discrete approach to segment the country road network using Open Street Map data is performed together with a segmentation of the country by municipalities followed by a map-matching algorithm used to map the accidents. The goal is to estimate how the accident count is affected by external variables like the weather conditions or the distance of the accident to its nearest intersection by applying a method of Poisson Regression. The results show the factors that influence more on the number of accidents are the normal weather and light conditions like clear day, daylight and dry surface; also, it is proved that the distance from the accident to its nearest intersection does not affect although being in an urban area does. All open codes developed for this thesis is available as notebooks.