Predicting Risky and Aggressive Driving Behavior among Taxi Drivers: Do Spatio-Temporal Attributes Matter?

Area of Science:

• Traffic safety research
• Data science applications in transportation

Background:

• Risky driving behaviors significantly increase traffic accident risk and severity.
• Existing statistical methods struggle to identify complex correlations and achieve high prediction accuracy for traffic violations.
• Taxi drivers represent a substantial portion of traffic violations in Luzhou, China.

Purpose of the Study:

• To comprehensively investigate traffic violations among taxi drivers in Luzhou, China.
• To apply spatial analysis and machine learning (ML) for understanding and predicting traffic violations.
• To compare the performance of different ML models in classifying violation types.

Main Methods:

• Utilized georeferenced taxi driver violation data from 2016.
• Performed descriptive analysis to summarize violation statistics.
• Employed Geographic Information System (GIS) for hotspot mapping of violations.
• Applied machine learning models: Decision Tree, AdaBoost, and a Stack Model for prediction and classification.

Main Results:

• Over-speeding was identified as the most frequent violation.
• Hotspot maps were generated to guide targeted interventions.
• Machine learning models demonstrated robustness in predicting violation types.
• The Stack Model achieved superior performance across evaluation metrics (accuracy, F1-score, specificity, log loss).

Conclusions:

• Machine learning, particularly the Stack Model, offers a powerful approach for predicting and classifying traffic violations.
• Spatial analysis and ML integration provide valuable insights for traffic safety management.
• Findings can inform targeted strategies to mitigate risky driving behaviors and reduce accidents.