Random forest models for motorcycle accident prediction using naturalistic driving based big data.

Area of Science:

• Traffic Safety
• Machine Learning Applications
• Big Data Analytics

Background:

• Traditional motorcycle accident studies rely on surveys and police reports, which may lack detailed behavioral and contextual data.
• Existing methods often overlook the predictive power of granular vehicle trajectory patterns.

Purpose of the Study:

• To develop a predictive model for motorcycle accident involvement using big data analytics.
• To identify key risk factors associated with motorcycle accidents, particularly those derived from trajectory data.
• To leverage machine learning for proactive identification of high-risk motorcyclists.

Main Methods:

• Utilized big data from GPS vehicle trajectory patterns, combined with self-reported accident data and rider characteristics.
• Employed a Random Forest machine learning algorithm to predict accident likelihood.
• Extracted and analyzed features including mobility patterns, acceleration events, aggressive overtaking events, and socio-economic factors.

Main Results:

• The developed model successfully predicts the likelihood of a motorcyclist's involvement in an accident.
• Aggressive overtaking event-based features were identified as having the highest impact on motorcycle accidents.
• Feature importance analysis highlighted specific trajectory-derived behaviors as critical risk indicators.

Conclusions:

• Machine learning models integrating big data, especially trajectory information, can effectively identify at-risk motorcyclists.
• Focusing safety interventions on behaviors like aggressive overtaking can significantly enhance motorcycle safety.
• This approach offers a data-driven strategy for reducing motorcycle accidents and improving rider safety.