Improving Crash Frequency Prediction from Simulated Traffic Conflicts Using Machine Learning Based Microsimulation
Researchers have explored the use of machine learning (ML) behavior models in traffic microsimulation to improve crash frequency prediction. By comparing ML models with traditional rule-based models at five intersections in Leeds, UK, they found that ML-generated conflicts aligned better with real-world crash data. However, directly using ML-simulated crashes for prediction proved inaccurate, indicating that while ML can realistically reproduce conflicts, it has not yet achieved realistic crash generation. AI
IMPACT Machine learning models show promise in enhancing traffic safety predictions by simulating more realistic conflict dynamics.