Traffic Incident Prediction Using Wavelet Based Feature Extraction and Artificial Neural Networks

The availability of huge traffic-related data enables us to evaluate and analyze the sources of traffic congestion and accidents in a systematic manner. Several researchers have explored ways to exploit the data for improving traffic safety and efficiency. Even though traffic accidents are impacted by several independent factors including road conditions, pavement surface criteria, human factors, lighting, weather conditions and traffic flow, the daily frequency of accidents can be treated as a set of signal data and additional analyses can be applied to predict the signal behavior. In this paper, an analytical approach presented that explores the prediction of accident frequencies using signal wavelet decomposition-based de-noising and then applying artificial neural network (ANN) for prediction of these features. A single iteration of the wavelet transform is applied to the accident frequency data to reduce the signal noise followed by implementation of four different training algorithms for the neural network prediction model. The accident data for the county of Los Angeles, collected between 2009 and 2013, were utilized to develop the proposed model. Daily frequency of accident data over the county region were extracted and combined into a single database and presented as a signal. This study did not focus on a specific geographic location or a highway segment for extraction of data. However, to increase the potential adoption of the results, a more detailed level of analysis should be performed in future studies. Three levels of wavelet decomposition were applied to the accident frequency data. The decomposed part of the signal was then used to train and test the neural network. The Levenberg-Marquardt training algorithm coupled with back-propagation of errors performed the best for training and validating the neural network prediction model. The preliminary results of this study were encouraging; however, it requires further investigation to improve the reliability of the prediction model.

• Supplemental Notes:
  • This paper was sponsored by TRB committee ANB20 Standing Committee on Safety Data, Analysis and Evaluation.
• Authors:
  • Cruz, Hector
  • Agarwal, Shaurya
  • Mazari, Mehran
• Conference:
  • Transportation Research Board 97th Annual Meeting
  • Location: Washington DC, United States
  • Date: 2018-1-7 to 2018-1-11
• Date: 2018

Language

• English

Media Info

• Media Type: Digital/other
• Features: Figures; References;
• Pagination: 14p

Subject/Index Terms

• TRT Terms: Crash causes; Crash data; Crash rates; Mathematical prediction; Neural networks
• Geographic Terms: Los Angeles (California)
• Subject Areas: Data and Information Technology; Highways; Safety and Human Factors;

Filing Info

• Accession Number: 01664155
• Record Type: Publication
• Report/Paper Numbers: 18-05331
• Files: TRIS, TRB, ATRI
• Created Date: Mar 26 2018 9:17AM