چكيده به لاتين
ABSTRACT
The warning growing trend of fatalities and injuries related to road accident has convinced researchers to focus on different types of traffic safety studies; however, developing accident prediction models probably has been the most frequent approach. In this regard, considerable energy has been made to find out among all well-known prediction modeling techniques, which one performs better, based on crash-related data. Furthermore, the present study aims to evaluate how crashes on the urban network can be affected by contributing factors such as hourly traffic flow characteristics. In this regard, four types of CI modeling technique, including MLP, ANFIS, Hybrid ANFIS-GA, and Hybrid ANFIS-PSO have been used to predict the accident frequencies; subsequently, their results were compared according to the study objectives. The dataset consists of three sets of variables, including a total of 1370 crash data, traffic flow at the moment of accident occurrence, and environmental conditions, such as highway geometry on roadway segments of Mashhad during the morning peak hour (i.e. 7:00- 8:00 A.M.) in the year 2014. According to the results, ANFIS-GA exhibited the best performance in forecasting accident frequency according to available data, while the MLP has been failed in the learning process. In addition, GA performed better in optimizing of ANFIS in comparison with PSO. Actually, not only did not using PSO improve the ANFIS performance, but also caused negative influence in its prediction accuracy. Furthermore, based on sensitivity analysis results, Speed, Ln (VKT) and width are the most significant variables; however, parking and median are not as significant as other variables.
Keywords: ANFIS, Nueral Network, MLP, Genetic Algoritm, PSO, Accident prediction models, Safty
