Road accidents analysis with bayesian method as a function of the infrastructural, environmental and roadway features

The aim of the study presented here is to divide a crash database into homogeneous clusters by using a Hard C-means algorithm to assess the expected number of crashes per year, with a known number of injuries and deaths over an 8-year period following the Empirical Bayes method. More than 4,500 km of road network located in Southern Italy was analyzed from 2003 to 2010, almost 404 km of which were freeways with 5,405 crashes, and 3,700km rural roads with 4,944 crashes and 400km of urban roads with 22,091 crashes. The explanatory variables used to split the crash database into homogeneous clusters derived from laborious computations and a careful study of the reports provided by the police as well as a survey on site at the time of the crash. The solution for correct clustering was an iterative optimization process of selective variables as follows: road category (freeway, rural, urban road), road scenario represented by a particular combination of infrastructure and environmental conditions (dry/wet road surface, day/night light condition, present or absent intersection), main crash types (head-on/side collisions, rear-end collisions, and single-vehicle run-off-road crashes), number of crashes, injuries and deaths, and human factors (percentage of drivers involved by gender). The best solution was obtained using 4 final clusters for freeways, 10 final clusters for rural roads, and 5 final clusters for urban roads. The Bayesian method was applied at each cluster: crash information for the period 2003-2010 was split so that seven years (2003 – 2009) were used as a reference base-population to specify the crash frequency per year, while the remaining one year (2010) was used to check the reliability of the prediction. On the basis of the reliable results, crash risk density maps are proposed to assess crash frequency per year when no crash trend is available.
Autori: 
Zilioniene D., Russo F., Busiello M., Dell’Acqua G.
Anno del prodotto: 
2013
Titolo convegno: 
The XXVIII International Baltic Road Conference
Luogo del Convegno: 
Vilnius
Pagina inizio: 
1
Pagina Fine: 
10
Lingua: 
English
Societa' Italiana Infrastrutture Viarie - S.I.I.V. - Cod.Fis. 93024730421