Traffic accident duration prediction using text mining and ensemble learning on expressways.

Area of Science:

• Traffic Safety and Management
• Data Science and Machine Learning

Background:

• Accurate prediction of traffic accident duration is crucial for enhancing road safety and mitigating congestion.
• Existing research has largely overlooked the potential of analyzing textual data from initial incident reports.

Purpose of the Study:

• To develop a predictive model for traffic accident duration by integrating text data fusion and ensemble learning.
• To identify significant textual features that influence accident duration.

Main Methods:

• A preprocessing scheme for accident duration text data was established.
• The random forest (RF) algorithm was employed for feature selection from textual data.
• Text feature vectors were utilized in various machine learning models, including Decision Tree, KNN, SVR, RF, GBDT, and XGBoost.

Main Results:

• The enhanced RF model demonstrated high prediction accuracy, evidenced by strong RMSE, MAPE, and R² values.
• Key textual factors influencing accident duration were identified.
• A cumulative importance of 60% for textual features was found to be sufficient for accurate traffic accident prediction.

Conclusions:

• Textual data analysis, combined with ensemble learning, offers a powerful approach to predicting traffic accident duration.
• The findings provide valuable insights for optimizing input data in text prediction models and reducing traffic congestion caused by accidents.