Evaluation of the level of responsibility in pedestrian crashes using machine learning algorithms

Area of Science:

• Road safety research
• Traffic accident analysis
• Machine learning applications

Background:

• Pedestrian traffic crashes cause significant casualties.
• Accurate responsibility attribution is vital for legal and policy decisions.
• Limited research exists on responsibility in pedestrian collisions.

Purpose of the Study:

• To evaluate supervised Machine Learning (ML) models for classifying driver and pedestrian responsibility in crashes.
• To identify key variables influencing responsibility attribution.
• To support judicial and police authorities in objective decision-making.

Main Methods:

• Utilized a dataset of 510 pedestrian crashes from Spanish police and judicial reports.
• Evaluated supervised classification models including Decision Trees (DT), Naïve Bayes (NB), and Support Vector Machine (SVM).
• Analyzed 14 binary variables across human, technological, structural, and normative subsystems.

Main Results:

• Decision Trees (DT) demonstrated superior performance compared to Naïve Bayes (NB) and Support Vector Machine (SVM).
• Possession of a driver's license was the most influential factor (47.26%) in determining responsibility.
• Other significant factors included pedestrian location (15.35%), driver impairment (7.24%), and distracted driving (7.04%).

Conclusions:

• Machine learning, specifically DT, offers an effective tool for objective responsibility attribution in pedestrian crashes.
• Identifying key influencing factors can inform road safety strategies and policy development.
• This approach enhances efficiency and objectivity for legal and police authorities.