View in EDS

A k-nearest text similarity-BiGRU approach for duration prediction of traffic accidents on expressways

Saved in:
Bibliographic Details
Title: A k-nearest text similarity-BiGRU approach for duration prediction of traffic accidents on expressways
Authors: Jiaona Chen, Jin Zhang, Peng Wang, Yinli Jin
Source: Discover Applied Sciences, Vol 7, Iss 7, Pp 1-21 (2025)
Publisher Information: Springer, 2025.
Publication Year: 2025
Collection: LCC:Science (General)
Subject Terms: Traffic safety, Traffic accident duration, Text similarity, Text vector, BiGRU, Science (General), Q1-390
Description: Abstract Accurate prediction of traffic accident duration is critical for alleviating congestion and optimizing traffic management. This study proposes a k-nearest text similarity-BiGRU(Bidirectional Gated Recurrent Unit) approach to predict the duration of traffic accidents using textual records. Firstly, text vectors are extracted through a TF-IDF (Term Frequency-Inverse Document Frequency) model to vectorize textual descriptions. Secondly, the text similarity is quantified based on the extracted text vectors. Therefore, the independent variables for predicting accident duration are obtained through a text similarity evaluation, which serves as the feature vector. Finally, BiGRU networks are utilized for the regression prediction of traffic accident duration using the constructed feature vectors. In addition, the proposed approach is evaluated by the RMSE (root mean square error) and MAPE (mean absolute percent error) utilizing survey information gathered from expressways of Shaanxi Province in China. A rigorous analysis of the prediction performances is compared with different regression methods. Experimental results demonstrate that the k-nearest text similarity-BiGRU outperforms other models. Moreover, the RMSE and MAPE of the TF-IDF model based on BiGRU are 47.77 and 43.74%, respectively. In general, the MAPE is much better than that of regression prediction models employed in previous studies. This work validates the feasibility of text-driven feature engineering for predicting the duration of traffic accidents, and provides a robust tool for traffic management systems.
Document Type: article
File Description: electronic resource
Language: English
ISSN: 3004-9261
Relation: https://doaj.org/toc/3004-9261
DOI: 10.1007/s42452-025-07366-7
Access URL: https://doaj.org/article/36efc89ea6e74a47bc67939e7d9e3f3d
Accession Number: edsdoj.36efc89ea6e74a47bc67939e7d9e3f3d
Database: Directory of Open Access Journals
Description
Abstract:Abstract Accurate prediction of traffic accident duration is critical for alleviating congestion and optimizing traffic management. This study proposes a k-nearest text similarity-BiGRU(Bidirectional Gated Recurrent Unit) approach to predict the duration of traffic accidents using textual records. Firstly, text vectors are extracted through a TF-IDF (Term Frequency-Inverse Document Frequency) model to vectorize textual descriptions. Secondly, the text similarity is quantified based on the extracted text vectors. Therefore, the independent variables for predicting accident duration are obtained through a text similarity evaluation, which serves as the feature vector. Finally, BiGRU networks are utilized for the regression prediction of traffic accident duration using the constructed feature vectors. In addition, the proposed approach is evaluated by the RMSE (root mean square error) and MAPE (mean absolute percent error) utilizing survey information gathered from expressways of Shaanxi Province in China. A rigorous analysis of the prediction performances is compared with different regression methods. Experimental results demonstrate that the k-nearest text similarity-BiGRU outperforms other models. Moreover, the RMSE and MAPE of the TF-IDF model based on BiGRU are 47.77 and 43.74%, respectively. In general, the MAPE is much better than that of regression prediction models employed in previous studies. This work validates the feasibility of text-driven feature engineering for predicting the duration of traffic accidents, and provides a robust tool for traffic management systems.
ISSN:30049261
DOI:10.1007/s42452-025-07366-7