Data-driven crash prediction by injury severity using a recurrent neural network model based on Keras framework.

Area of Science:

• Traffic Safety
• Data Science
• Machine Learning

Background:

• Big data and deep learning technologies are increasingly utilized in data-driven applications.
• Predicting traffic injury severity is crucial for urban safety and resource allocation.

Purpose of the Study:

• To propose a recurrent neural network (RNN) model for predicting urban traffic injury severity.
• To leverage clustering and deep learning for enhanced prediction accuracy and speed.

Main Methods:

• Utilized OPTICS clustering algorithm on Nevada crash data (2014-2017) for Las Vegas.
• Developed a recurrent neural network (RNN) model using the Keras framework.
• Optimized model parameters including loss function, activation function, and optimizer.

Main Results:

• The RNN model achieved high accuracy in predicting injury severity.
• The model demonstrated a high training speed, suitable for real-time applications.
• Successful visualization of model training results was achieved.

Conclusions:

• The proposed data-driven RNN model offers an effective method for predicting urban traffic injury severity.
• The findings provide potential insights for improving road safety strategies.
• This approach demonstrates the power of machine learning in analyzing complex urban safety data.