PRIMENA HIJERARHIJSKIH MODELA ZA PREDIKCIJU SAOBRAĆAJNIH NEZGODA U URBANIM SREDINAMA

Pljakić, Miloš; Jovanović, Dragan; Matović, Boško; Bačkalić, Svetlana; Mićić, Spasoje

A04

PRIMENA HIJERARHIJSKIH MODELA ZA PREDIKCIJU SAOBRAĆAJNIH NEZGODA U URBANIM SREDINAMA

APPLICATION OF THE HIERARCHICAL BAYES METHODS FOR THE PREDICTION AND RANKING OF AGGREGATE ACCIDENTS ON THE TRAFFIC ANALYSIS ZONE ON THE URBAN AREAS OF THE CITY OF NOVI SAD

Miloš Pljakić, Dragan Jovanović, Boško Matović, Svetlana Bačkalić, Spasoje Mićić

Pljakić, M., Jovanović, D., Matović, B., Bačkalić, S., Mićić, S., (2019). PRIMENA HIJERARHIJSKIH MODELA ZA PREDIKCIJU SAOBRAĆAJNIH NEZGODA U URBANIM SREDINAMA. 14 th International Conference - Road Safety in Local Communities, .

Rezime: Hijerarhijski modeli pružaju pomoć istraživačima, koji se susreću sa različitim analitičkim ograničenjima, kako bi omogućili razvoj pouzdanih prediktivnih modela i istražili određene faktore koji utiču na frekvenciju saobraćajnih nezgoda. Cilj ovog rada je primena prediktivnih modela kako bi se identifikovali faktori koji utiču na frekvenciju saobraćajnih nezgoda u okviru saobraćajnih zona na području grada Novog Sada. Modeli koji su primenjeni nad prikazanim setom podataka su Bajesov višestepeni Poasonov model kao i Bajesov višestepeni negativni binomni model. Na osnovu kriterijumu za odabir modela, najbolje performanse ima Bajesov višestepeni negativni binomni model, gde su predstavljeni svi uticajni faktori. Rezultati ovog istraživanja mogu biti od velike koristi za lokalnu zajednicu u cilju planiranja preventivnih mera koje je potrebno sprovesti na određenoj lokaciji.

Ključne reči: Hijerarhijski modeli, Saobraćajne zone, Novi Sad.

Abstract: Hierarchical Bayes models helped researchers which have computational constraints and allow researchers and practitioners to develop more realistic prediction models and they are examining factors which influence on the frequency road accidents. The objective of this paper is to infer the process of listing the most dangerous traffic analysis zones, based on the available accident data. The hierarchical random effects model allows the specification of different sources of variation, namely the variation between zones and the variation within each zone. The Gibbs sampler is used to explore the distribution of the accident proportions, this approach showed better estimate coefficients than Maximum Likelihood Estimate (MLE). An important advantage of the Gibbs sampler is the possibility to sample complex functions of the accident proportions, like the rank of zones. It is shown that the ranking itself could be seen as a density accident in the only zone. Since the accident proportions have a stochastic character, the ranking of zones based on the mean posterior proportion cannot be deterministic. In this paper, Bayesian hierarchical modeling techniques are used to identify and rank hazardous traffic analysis zones in Novi Sad. The research period relates to three years (2015-2017), where various types of traffic accidents are covered. This paper investigates the question whether a ranking alone can give enough evidence for the selection of dangerous zones on the urban areas of the City of Novi Sad.

Keywords: Predictive modelling, Accidents, Traffic Analysis Zone, Urban areas.

Presented by: Miloš Pljakić