Abstract
Every year, traffic accidents cause a great deal of death, serious injury, and financial damage, making it a major global problem. Predicting accident severity is crucial for implementing effective traffic management and safety strategies. This study employs Random Forest and Deep Neural Network (DNN) models and so on from machine learning algorithms to predict the traffic accidents severity, identifying related outcomes from a joint dataset collected by the Seattle Department of Transportation as well as UKs Department for Transport. The results showcase that snowy road surfaces, standing water and dusk or insufficient lighting conditions causes respectively an excess of 10.75%, 10.44% and 13.01% above the average number of causalities in traffic accidents. The DNN model achieved the highest accuracy (91.12%), outperforming other methods by 310 percentage, especially in severe crash detection performance. In contrast, Random Forest model offers better stability across multiple classification threshold, illustrating better performance on general tasks. By diagnosing the high-risk conditions, traffic authorities can implement tailored interventions, ranging from boosting road maintenance efforts during adverse weather, improving street lighting, and increasing safety measures during peak times, to boost the standard of protection. Future research should investigate the method to optimise the robustness and generalization of DNN model through techniques like threshold adjustment, or adopting ensemble learning mechanisms as well as devising new segments that are more relevant.
Published Version (Free)
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have