Road Traffic Accident Risk Prediction and Key Factor Identification Framework Based on Explainable Deep Learning

Aim:

     The aim of this study is to develop a robust and accurate traffic accident risk prediction model by leveraging deep learning techniques such as CNN (Convolutional Neural Network), BiLSTM (Bi-directional Long Short-Term Memory), and GRU (Gated Recurrent Unit) models.

Abstract

      Traffic accidents are a significant concern globally, and accurately predicting their risks can play a crucial role in preventing accidents and ensuring road safety. This study builds upon existing models by integrating CNN, BiLSTM, and GRU models to predict traffic accident risks using the US dataset. A novel combination of spatial and temporal feature extraction methods is employed to enhance the predictive performance of the system. Through comprehensive experiments, the model demonstrates improved accuracy compared to previous studies, achieving superior prediction results. The user-friendly interface, developed using Flask, allows for efficient visualization and interaction with the prediction results. The study aims to contribute to better-informed traffic safety policies and improved travel route planning for road users.

Existing System

       Existing systems for traffic accident risk prediction primarily focus on predicting accident risks through traditional machine learning models like logistic regression, decision trees, and random forests. These models, although effective, often fail to capture the complex spatiotemporal dynamics of traffic accidents. While deep learning models such as CNN, LSTM, and BiLSTM have shown promise in improving prediction accuracy, most existing studies focus only on specific datasets or fail to integrate multiple deep learning techniques. Additionally, many existing systems lack an intuitive user interface, which limits their practical application for road users and safety authorities.

Problem Definition

       The primary challenge lies in accurately predicting traffic accident risks by considering both spatial and temporal features. Traditional machine learning models often fail to account for the complex interactions between different contributing factors such as traffic density, weather conditions, road types, and time of occurrence. Furthermore, the lack of an interactive and user-friendly interface for displaying prediction results hinders real-time decision-making.

 Proposed System

          The proposed system integrates deep learning techniques, including CNN, BiLSTM, and GRU models, to effectively predict traffic accident risks. By utilizing CNN for spatial feature extraction, BiLSTM for capturing temporal dependencies, and GRU for modeling sequential patterns in the data, the system provides a more comprehensive and accurate risk prediction. The model is trained on a US traffic accident dataset containing features like traffic density, weather conditions, accident severity, and time of occurrence. Additionally, a user-friendly interface developed using Flask allows users to input traffic data and receive real-time predictions, as well as visualize contributing factors affecting accident severity.

Advantage

• Improved accuracy in predicting traffic accident risks using deep learning techniques.
• The integration of CNN, BiLSTM, and GRU enables better modeling of spatiotemporal features.
• A user-friendly interface developed with Flask for easy access to prediction results.
• Real-time prediction capabilities help in making timely and informed decisions.
• Enhanced understanding of key contributing factors to traffic accidents.