A Bayesian network model to predict accidents on Swiss highways

Abstract

Although in 2014, Switzerland had an average of less than two fatalities per billion vehicle-kilometres, making its roads among the safest in Europe, still more than 17 000 traffic accidents occurred on Swiss communal roads, cantonal roads and national highways. On the highway network of approximately 1800 km alone, there were almost 1700 accidents involving personal injuries. In order to further reduce this number of accidents, it is important that accident risks are assessed as accurately as possible. A state-of-the-art methodology is used to develop a Bayesian probabilistic network model to estimate the number of accidents involving personal injury on the Swiss highway network. The developed model predicts the number of accidents on a given highway segment and can be used to identify segments with a high expected number of accidents. During validation, the number of accidents was correctly predicted on 86% of the segments with a tolerance of 25%. The model can also be used to conduct parametric studies, which help to ensure that the risk reduction interventions are effective and efficient. Road traffic and road infrastructure engineers and managers can use the model during the decision-making processes in the planning, construction and maintenance of road networks.