Injury Accident Frequency Forecasting Model in 4-Armed Signalized Intersections

Traffic accidents in urban signalized intersections cause great human and economic losses. Intersections especially in urban areas are one of the most critical hazardous locations with high risk of accidents. But in Iran, it does not seem any research work being implemented for studying safety at intersections. Therefore, optimum allocation of scarce resources for intersection safety promotion is difficult in this situation.The purpose of this study is to develop a statistical model to predict the frequency of injury accidents at 4-armed intersections and to diagnose contributory attributes which affect the accident frequency occurrences. Then, the ultimate purpose of the study would be to find cost-effective remedial measures to reduce the injury accidents at such intersections. In previous studies for intersection safety modeling, both poison and negative binomial distributions have been used. Important limitation related to poison distribution is that the ratio of variance to mean is one, which rarely is justified for traffic accident data. To compensate this limitation, negative binomial distribution is used. In this study a statistical forecasting model based on negative binomial distribution for injury accident frequency in 4-armed signalized intersections in the city of Mashhad, is developed. To obtain a reliable model, it is necessary that independent variables would be true representatives of real geometric, traffic and control attributes of the intersection an its approaches. These include the number of traffic lanes in each approach, the width of traffic lanes, existence of protected left and/or right turn lane(s) and its (their) number, width and length, the existence of median, the distance of bus-stop(s) to intersection, one-way or two-way the approaches, the angle between major and minor approaches, the number of phases in each cycle, the existence of control camera(s), Type of intersection control system, the volumes of left and right turnings and the total of traffic volume. Most of geometry related variables were taken from 1:2000 scales Mashhad map, and traffic related variables were obtained from Mashhad transportation and traffic organization. From 110 existed 4-armed intersections in the city of Mashhad, 50 of them were selected for this analysis, and related variables for the Iranian year 1385 (21st March 2006 – 20th March 2007) were collected. The statistical package Genmod in software SAS version 9.0 was used and the suitability of negative binomial distribution based on Deviance and Pearson x2 tests was controlled.The developed injury frequency accident model is as following: The model parameters are fully introduced in the paper manuscript as following:Y`TIA= Total of predicted Injury Accidents at the Intersection, xNL= Number of traffic Lanes at the intersection approaches, xM= the existence of Median at the intersection, xNP= Number of phases per cycle, xST =Type of control system intersection, xSC= the existence of Surveillance Camera at the intersection, and xTV= Total traffic Volume at the intersection approaches. To calibrate the developed model and to examine its suitability, R2 was calculated. Validation of the model, by using a new set of data and employing paired samples t-test method was controlled. Developed model showed a strong relationship exists between frequency of injury accidents and independent variables of road geometry, the type of control system and traffic characteristics. The results of this study show that six of independent variables considerably affect the safety of signalized intersections. These are: approach lane numbers, the existence of median, number of phases per cycle, Type of intersection control system, control camera and approach traffic volume.