TRAFFIC COLLISION TIME SERIES ANALYSIS (A CASE STUDY OF KARAJ–QAZVIN FREEWAY)

Abstract. The application of Auto-correlation Function (ACF) and Partial Auto-correlation Function (PACF) in recent years has been improved in analyzing big traffic data, modelling traffic collisions and decreasing processing time in finding collision patterns. Accident prediction models for short and long time can help in designing and programming traffic plans and decreasing road accidents. Based on the above details, in this paper, the Karaj-Qazvin highway accident data (1097 samples) and its patterns from 2009 to 2013 have been analyzed using time series methods.

In the first step, using auto correlation function (ACF) and partial auto correlation function (PACF), the rank of time series model supposed to be autoregressive (AR) model and in the second stage, its coefficients were found. In order to extract the accident data, ArcGIS software was run. Furthermore, MATLAB software was used to find the model rank and its coefficients. In addition, Stata SE software was used for statistical analysis. The simulation results showed that on the weekly scale, based on the trend and periodic pattern of data, the model type and rank, ACF and PACF values, an accurate weekly hybrid model (time series and PACF) of an accident can be created. Based on simulation results, the investigated model predicts the number of accident using two prior week data with the Root Mean Square Error (RMSE) equal to three.