inTformer: A Time-Embedded Attention-Based Transformer for Crash Likelihood Prediction at Intersections Using Connected Vehicle Data

  • 2023-08-29 16:51:05
  • B M Tazbiul Hassan Anik, Zubayer Islam, Mohamed Abdel-Aty
  • 0

Abstract

The real-time crash likelihood prediction model is an essential component ofthe proactive traffic safety management system. Over the years, numerousstudies have attempted to construct a crash likelihood prediction model inorder to enhance traffic safety, but mostly on freeways. In the majority of theexisting studies, researchers have primarily employed a deep learning-basedframework to identify crash potential. Lately, Transformer has emerged as apotential deep neural network that fundamentally operates throughattention-based mechanisms. Transformer has several functional benefits overextant deep learning models such as LSTM, CNN, etc. Firstly, Transformer canreadily handle long-term dependencies in a data sequence. Secondly,Transformers can parallelly process all elements in a data sequence duringtraining. Finally, a Transformer does not have the vanishing gradient issue.Realizing the immense possibility of Transformers, this paper proposesinTersection-Transformer (inTformer), a time-embedded attention-basedTransformer model that can effectively predict intersection crash likelihood inreal-time. The proposed model was evaluated using connected vehicle dataextracted from Signal Analytics Platform. Acknowledging the complex trafficoperation mechanism at intersection, this study developed zone-specific modelsby dividing the intersection region into two distinct zones:within-intersection and approach zone. The best inTformer models in'within-intersection,' and 'approach' zone achieved a sensitivity of 73%, and70%, respectively. The zone-level models were also compared to earlier studieson crash likelihood prediction at intersections and with several establisheddeep learning models trained on the same connected vehicle dataset.

 

Quick Read (beta)

loading the full paper ...