Application of Principal Component Analysis as a Variables Reduction Technique in Freeway Accident Prediction Models (A Case Study)

Traffic accidents are one of the most substantial concerns that call for improving road safety. In this paper, research work was performed to recognize the factors affecting crash frequency in rural freeway. Therefore crash data of Tehran-Ghom freeway were used as a case study. In this research, artificial neural network and Log-Normal model were proposed to estimate the number of road accidents in Tehran-Qom freeway. Average daily traffic volume, percentage of heavy vehicle, average speed and environmental effects were considered as independent variables. Thirty four-month period of accident data and parameters, were collected to use in analytical process of modeling and validation of them. To address influences of the main contributing factors on accident frequency, Principal Component Analysis (PCA) and Factor Analysis (PFA) techniques appeared to be useful l in analyzing variables in order to identify the most significant in accident-prediction model. Also sampling adequacy was measured by the Kaiser-Meyer-Olkin (KMO) and Bartlet test statistics in PCA technique. With the aim of evaluating efficiency of artificial neural network model against Log-Normal model in crashes modeling on rural freeways, these models were compared and the results revealed that artificial neural network model was more capable to estimate the number of road accidents in freeways. Results also showed that average speed of vehicles and average daily traffic volume were the most effective parameters in freeway accidents.