Developing a Graph Convolutional Neural Network in Crash Severity Modeling

Articles in Press

Document Type : Scientific - Research

Authors

1 Department of transportation, faculty of civil engineering, Iran university of science and technology, Tehran, Iran

2 Department of Transportation, Faculty of Civil Engineering, IUST

Abstract

Due to high socio-economical costs of traffic crashes, accident injury prediction is an important aspect in safety related studies. Additional to statistical methods that have been traditionally developed in crash severity modeling, some machine learning algorithms have also been employed in recent years. Neural Network being one such algorithm, have gained recognition in the safety related fields in recent field. However, as the distribution of crash data might be nonlinear and due to a number of complications, like inability to model unobserved heterogeneity, using the crash data in its original form for neural networks, may not yield the best result. So, the current study, inspired by the KNN algorithm, develops a hidden graph structure on the data, and then using Graph Convolutional Neural Networks, tries to extract more meaningful relations between crash severity and other explanatory variable. Different metrics adopted for the current task show that the performance of the proposed model is significantly improved over other machine learning models.

Keywords

Main Subjects

Quarterly Journal of Transportation Engineering

Articles in Press, Accepted Manuscript
Available Online from 11 October 2023

Files

History

  • Receive Date: 01 October 2023
  • Revise Date: 09 October 2023
  • Accept Date: 10 October 2023

Share

How to cite

Statistics