| Abstrakt: |
Road traffic accidents are among the leading causes of death in the world, which not only produce casualties, but also result in large financial losses. Discovering patterns based on which crashes occur under various circumstances, as well as recognizing the involved factors could play an important role in reducing the crash counts. Most researches so far carried out to predict the accidents occurred, have analyzed a limited number of similar roads, depending on the presented traffic data and models of road accident, which cannot be generalized for all roads with different characteristics. The model developed in this study, evaluates different types of factors influencing road crashes, rather than using only traffic data. For this purpose, as the first contribution of the present study, a comprehensive dataset has been created by merging and integrating different sources of data. These data include traffic data, climatic conditions, different road characteristics, geographical information, roads lighting, census data and regional demographics. The presented model investigates the road traffic accidents occurred in thousands of miles of highways and priority roads of United Kingdom (UK), during 2013 to 2017. Using the constructed dataset, the factors influencing road accidents are identified through various methods; and consequently, a CNN-based classification method for road accident prediction is implemented and compared with its counterparts. The results of this study suggest that the traffic features play the most significant role in crash occurrence, compared to other available components of the dataset. |
Full text from SpringerLink