A model development on GIS-driven data to predict temporal daily collision through integrating Discrete Wavelet Transform (DWT) and Artificial Neural Network (ANN) algorithms; case study: Tehran-Qazvin freeway

Abstract

The aim of this study is to develop a model to predict temporal daily collision by integrating of Discrete Wavelet Transform (DWT) and Artificial Neural Network (ANN) algorithms. As a case study, the integrated model was tested on 1097 daily traffic collisions data of Karaj-Qazvin freeway from 2009 to 2013 and the results were compared with the conventional ANN prediction model. In this method, initially, the raw collision data were analyzed, normalized, and classified via Geographical Information System (GIS). Partial Autocorrelation Function (PACF) was also utilized to evaluate the temporal autocorrelation for consecutive existing daily data. The results of this study showed that the proposed integrated DWT-ANN method provided higher predictive accuracy in daily traffic collision than ANN model by increasing coefficient of determination (R²) from 0.66 to 0.82.

Keywords:

• Discrete Wavelet Transform
• Artificial Neural Network
• traffic collision
• GIS
• Partial Autocorrelation Coefficient

Acknowledgements

We gratefully acknowledge the contributions of Mohammad Salamati, North Carolina State University, and Abdul Reza Alavi Qarabagh, Azad University of Shahrood, for their technical assistance. This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.

Disclosure statement

No potential conflict of interest was reported by the author(s).