A random forest and SHAP-based analysis of motorcycle crash severity in Thailand: Urban-rural and day-night perspectives
Road traffic crashes pose significant public safety concern globally, causing severe injuries and fatalities. Motorcyclists face heightened crash risks and injury severity, particularly in developing countries like Thailand, where motorcycles serve as a primary mode of transportation. This study exa...
Saved in:
| Main Authors: | , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Elsevier
2025-09-01
|
| Series: | Transportation Engineering |
| Subjects: | |
| Online Access: | http://www.sciencedirect.com/science/article/pii/S2666691X25000685 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Road traffic crashes pose significant public safety concern globally, causing severe injuries and fatalities. Motorcyclists face heightened crash risks and injury severity, particularly in developing countries like Thailand, where motorcycles serve as a primary mode of transportation. This study examines motorcycle crash severity across four distinct scenarios: urban daytime, urban nighttime, rural daytime, and rural nighttime. Analyzing 12,266 crashes from Thailand's Highway Accident Information Management System (HAIMS) spanning 2020–2022, Random Forest (RF) modeling combined with SHapley Additive exPlanations (SHAP) was applied to identify key severity determinants while enhancing model interpretability. The analysis revealed significant variations across scenarios based on roadway characteristics, environmental conditions, crash causes, and vehicle involvement. Crashes involving large trucks, head-on collisions, roads with depressed medians, and darkness were associated with increased severity. Conversely, those involving passenger cars, side-swipe collisions, roads with barrier medians, and well-lit locations exhibited lower severity. To assess its effectiveness, RF was benchmarked against Logistic Regression and Decision Tree models and consistently outperformed them across all crash scenarios. The models achieved classification accuracies of 66.5 % (urban day), 64.7 % (urban night), 63.8 % (rural day), and 65.9 % (rural night), while SHAP analysis illuminated the factors driving these predictions. These findings offer critical insights for policymakers and transportation planners, enabling the development of targeted interventions tailored to specific environmental and temporal conditions. By integrating machine learning with explainable artificial intelligence, this study advances data-driven approaches for enhancing motorcycle safety and crash prevention measures. |
|---|---|
| ISSN: | 2666-691X |