Evaluating Road Crash Severity Prediction with Balanced Ensemble Models
This study evaluates the performance of an ensemble of five ML models (Random Forest, XGBoost, AdaBoost, LightGBM and CatBoost) on crash data from New South Wales, Australia. The model is evaluated based on ROC-AUC score, with a result of 0.68, indicating a moderate level of predictive accuracy. Fea...
Saved in:
| Main Author: | |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Findings Press
2024-04-01
|
| Series: | Findings |
| Online Access: | https://doi.org/10.32866/001c.116820 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | This study evaluates the performance of an ensemble of five ML models (Random Forest, XGBoost, AdaBoost, LightGBM and CatBoost) on crash data from New South Wales, Australia. The model is evaluated based on ROC-AUC score, with a result of 0.68, indicating a moderate level of predictive accuracy. Feature importance analysis reveals the key predictors being the vehicle type involved, with sedans/hatchbacks and motorcycles being the most common in fatal crashes, and the collision type, with vehicle-to-object impacts often leading to fatalities. Furthermore, fatal crashes occur more on Saturdays, in country non-urban LGAs and speed limits of 100 km/h as the most usual settings for fatal accidents. |
|---|---|
| ISSN: | 2652-8800 |