Citation

Zhao HZ, Rao G, Qiu H, Zhang C, Fei Z, Qiu M, Kung SY. Lect. Notes Comput. Sci. 2021; 12817: 41-52.

Copyright

(Copyright © 2021, Holtzbrinck Springer Nature Publishing Group)

DOI

10.1007/978-3-030-82153-1_4

PMID

unavailable

Abstract

Road traffic accident prediction has always been a complex problem for intelligent transportation since it is affected by many factors. However, to simplify the calculation complexity, most of the current research considers the impact of a few key factors and ignores multiple factors' impact in reality. To address this problem, we propose traffic accident prediction methods based on multi-factor models. The model introduces information including the severity of the traffic accident, the weather in which the accident occurred, and the external geographic environment to construct a multiple factors model to improve the prediction accuracy. Also, we can use more factors to construct the multi-factor model with the enrichment of data information. The multi-factor model can overcome the shortcomings of existing models in filtering data fluctuations and achieve more accurate predictions by extracting time-periodic features in time series. Furthermore, we combine the multi-factor models with different deep learning models to propose multiple traffic accident prediction methods to explore multi-factor models' effects in traffic accident prediction. The experimental results on the 2004-2018 Connecticut Crash Date Repository data of the University of Connecticut show that the

Keywords

Bi-LSTM-attention; Multi-factor model; Road traffic safety; Time series; Traffic accident prediction