مهدي هادي نيا

Road accidents, especially on interurban roads, are one of the main challenges of traffic safety. Identifying the factors that affect the severity an‎d patterns of accidents is of great importance for planning preventative interventions an‎d reducing crashes. According to statistics from the countryʹs forensic medicine organization, over 20,000 of our compatriots lose their lives annually in intercity road accidents. Therefore, this study aims to cluster highway police stations based on the characteristics of accidents that occurred within their jurisdiction. It uses a three-stage statistical an‎d machine learning model to find accident patterns in Isfahan province, a central province of Iran with over 12,000 km of intercity an‎d connecting roads. The study uses a database of intercity accidents in Isfahan province from 2017 to 2021, which was collected by the traffic police. In the first stage, a binary logistic regression with an odds ratio is used to identify the factors affecting the severity of accidents. In the second stage, the police stations are classified into three groups using hierarchical clustering based on the aggregated characteristics of accidents belonging to each station. In the final stage, different patterns at the highway police stations are identified using the K-Modes algorithm an‎d the Elbow method to determine the optimal number of clusters. This stage is performed only on accidents that, in the first stage (the accident severity model), had a higher probability of being fatal. Finally, this three-stage model was compared with a two-stage model. The results of the three-stage model led to the identification of seven distinct patterns of nighttime accidents that occurred at highway police stations in Isfahan province. Each cluster showed its own specific characteristics, such as the type of road, type of collision, an‎d age of the at-fault party. A comparison of the three-stage an‎d two-stage models showed that both models identified similar an‎d different patterns from each other an‎d can complement each other in providing the dominant patterns of road accidents. The results of this study also showed that each highway police station has its own specific type of accidents, so that each station can reduce the total number of accidents nationwide by implementing specific strategies an‎d consequently reducing accidents in its jurisdiction.

شدت تصادفات , خوشه‌بندي تصادفات , پاسگاه پليس‌راه , مدل سه‌مرحله‌اي , مدل كا-مدز

Crash severity , Crash clustering model , Road police station , Three-stage modeling , K-Modes

Mahdi Hadinia

Ali Tavakoli Kashani