This study aims to improve the accuracy and interpretability of traffic accident severity nowcasting by introducing a stacked Recurrent Neural Network (RNN) deep learning model. Accurately predicting traffic accident severity is crucial for enhancing traffic management and reducing the impact of accidents. We employed a stacked Bidirectional Gated Recurrent Unit (GRU) - Long Short Term Memory (LSTM) model with an attention mechanism, integrating multivariate accident data to capture complex temporal dynamics. The use of SHapley Additive exPlanations (SHAP) values enhances the interpretability of the model. The model demonstrates high reliability and effectiveness, achieving an accuracy of 88.06% and an F1-score of 0.867 in real-time applications. It provides valuable insights into the factors influencing predictions, making the decision-making process transparent. This framework not only advances predictive performance but also aligns with ethical AI deployment, making it a valuable tool for traffic management and policy formulation.
Explainable Traffic Accident Severity Prediction with Attention-Enhanced Bidirectional GRU-LSTM
2024 IEEE Smart World Congress (SWC) ; 1083-1090
2024-12-02
620927 byte
Conference paper
Electronic Resource
English
Comparison of traffic accident injury severity prediction models with explainable machine learning
Taylor & Francis Verlag | 2023
|Fusion attention mechanism bidirectional LSTM for short-term traffic flow prediction
Taylor & Francis Verlag | 2024
|Unidirectional and Bidirectional LSTM Models for Short-Term Traffic Prediction
DOAJ | 2021
|