Improving traffic accident severity prediction using MobileNet transfer learning model and SHAP XAI technique

Area of Science:

• Road safety
• Traffic accident analysis
• Predictive modeling

Background:

• Traffic accidents are a major cause of fatalities and injuries.
• Predictive modeling offers insights into accident factors.
• Lack of transparency in complex models hinders trust.

Purpose of the Study:

• To develop predictive models for accident severity using transfer learning.
• To identify key factors influencing accident prediction using Shapley values.
• To enhance trust and understanding of machine learning in road safety.

Main Methods:

• Employed transfer learning techniques including Multilayer Perceptron (MLP), Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM), Residual Networks (ResNet), EfficientNetB4, InceptionV3, Extreme Inception (Xception), and MobileNet.
• Utilized Shapley values to interpret model predictions and identify influential features.
• Focused on predicting the severity of injuries in traffic accidents.

Main Results:

• MobileNet model achieved the highest prediction accuracy at 98.17%.
• Shapley values provided insights into the impact of various features on accident prediction.
• Demonstrated the effectiveness of deep learning in analyzing traffic accident data.

Conclusions:

• Deep learning models, particularly MobileNet, are highly effective for predicting traffic accident severity.
• Interpretable AI methods like Shapley values are crucial for understanding and trusting predictive models.
• Findings can inform the development of targeted interventions to improve road safety.