Khishdari Abolfazl, Fallah Tafti Mehdi
a Road and Transportation Engineering, Civil Engineering Department , Yazd University , Yazd , Iran.
b Civil Engineering Department, Faculty of Engineering , Yazd University , Yazd , Iran.
Int J Inj Contr Saf Promot. 2017 Dec;24(4):519-533. doi: 10.1080/17457300.2016.1278237. Epub 2017 Jan 24.
The merits for development and application of crash frequency prediction models for safety promotion on any road type, with a focus on urban collector streets, are presented in this article. The city of Yazd, a medium-sized city in the middle of Iran, was selected as a case study and the data required for modelling crash frequencies along five collector streets comprising 31 street sections were collected. Six models including Poisson and negative binomial models and their deviations along with a hybrid artificial neural networks (ANN) model were developed to predict crash frequency along each street section. The overfitting problem was addressed using appropriate sensitivity analysis methods which were also used to identify the input variables with significant impact on the model performance. The results indicated that the developed hybrid ANN model provided the best performance in terms of accuracy and the number of input variables. The application of hybrid ANN model to evaluate the safety impacts of four different strategies, each resembled by one of the input variables of this model, indicated that these models can successfully be used for this purpose.
本文介绍了开发和应用碰撞频率预测模型以促进各类道路安全的优点，重点关注城市支路。伊朗中部的中等城市亚兹德市被选为案例研究对象，收集了五条包含31个路段的支路沿线碰撞频率建模所需的数据。开发了六个模型，包括泊松模型和负二项式模型及其偏差，以及一个混合人工神经网络（ANN）模型，以预测每个路段的碰撞频率。使用适当的敏感性分析方法解决了过拟合问题，这些方法还用于识别对模型性能有重大影响的输入变量。结果表明，所开发的混合ANN模型在准确性和输入变量数量方面表现最佳。将混合ANN模型应用于评估四种不同策略的安全影响，每种策略由该模型的一个输入变量表示，结果表明这些模型可成功用于此目的。
