Traffic accidents are a major public safety issue, causing widespread harm and economic losses due to factors like human error, adverse weather, and vehicle malfunctions. Unintentional collisions between cars or between cars and pedestrians are known as traffic accidents, and they frequently cause serious injuries, fatalities, and substantial property damage. Usually, bad weather, inadequate road infrastructure, distracted or intoxicated driving, and driver error are the causes of these events. Drivers must follow safety precautions to reduce these dangers, such as following traffic laws, keeping their cars in good condition, wearing seat belts and helmets, and avoiding distractions while operating a motor vehicle. Predicting traffic accidents accurately is essential for putting preventative safety measures into place, lessening the demand on emergency services, and eventually saving lives. In order to predict the severity of traffic accidents, we provide a deep learning-driven method in this study that blends Convolutional Neural Networks (CNNs) with conventional machine learning models, such as Random Forest (RF), Logistic Regression (LR), and XGBoost. High-level spatial features are extracted from raw traffic data using CNNs and then fed into machine learning classifiers to provide precise predictions. According to our experimental findings, the CNN + XGBoost model outperformed CNN + RF (93%) and CNN + LR (92%), achieving an accuracy of 95%. This integrated strategy demonstrates how deep learning and machine learning may be combined to improve prediction performance and offer insightful information for managing traffic safety.
A Deep Learning-Driven Approach to Predict Traffic Accident Severity
2025-05-02
399736 byte
Conference paper
Electronic Resource
English
Traffic accident severity prediction based on interpretable deep learning model
Taylor & Francis Verlag | 2025
|Classification of Traffic Accident Severity Using Machine Learning Models
Springer Verlag | 2025
|