Investigating Hit-and-Run Crashes with Gradient Boosting Decision Trees
Author Information
Author(s): Han Baorui, Huang Haibo, Li Gen, Jiang Chenming, Yang Zhen, Zhu Zhenjun
Primary Institution: School of Automobile and Traffic Engineering, Nanjing Forestry University, Nanjing, Jiangsu, China
Hypothesis
What factors contribute to a perpetrator’s escape behavior in hit-and-run crashes?
Conclusion
The study found that the Gradient Boosting Decision Tree model outperforms traditional methods in predicting hit-and-run crashes.
Supporting Evidence
- The GBDT model achieved the lowest negative log-likelihood of 0.282.
- The misclassification rate for the GBDT model was 0.096, the lowest among compared models.
- The AUC for the GBDT model was 0.803, indicating strong predictive performance.
- Hit-and-run crashes accounted for 10.97% of the total crash accidents in the dataset.
Takeaway
This study used a smart computer model to figure out why some drivers run away after accidents, helping to make roads safer.
Methodology
The study used a Gradient Boosting Decision Tree model trained on data from the U.S. Crash Report Sampling System.
Limitations
The study did not include psychological factors affecting driver behavior and was limited to specific geographical regions.
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website