Graphical-Based Neural Network Model for Determining the Level of Roadway Breakdown Injuries.

Predicting injury severity in road accidents is vital for improving transportation safety and emergency response. Traditional methods using factors like weather, vehicle type, and road conditions often fail to capture complex interactions. This study presents a Graph Neural Network (GNN) model that treats crash data as a graph, revealing intricate relationships among influencing factors. The model was tested against XGBoost, Random Forest, and Artificial Neural Networks using UK transport data. Uniform preprocessing ensured fairness across models. Evaluation used metrics like accuracy, precision, recall, F1-score, and MCC. GNN achieved the highest accuracy at 85.55%, outperforming XGBoost (83.36%), RF (83.18%), and ANN (83.27%). Its advantage lies in modeling relational patterns more effectively. The results demonstrate GNN’s potential in real-world traffic systems for more accurate injury predictions.