Road Traffic Accident Risk Prediction System Based on CatBoost

Project Basic Information

• Original Author/Maintainer: Aryan Deo
• Source Platform: GitHub
• Original Project Title: Road-Accident-Risk-Prediction-using-CatBoost-Regressor
• Release/Update Date: June 10, 2026

Core Content This project uses CatBoost Regressor to build a machine learning model that predicts traffic accident risk scores (0.0-1.0) based on multi-dimensional factors including road conditions, weather, lighting, traffic, and speed limits. The model achieves an R² score of 0.8855, providing a scientific tool for road safety management.

Road traffic accidents are one of the main causes of casualties and property losses worldwide. According to the World Health Organization, about 1.3 million people die from road traffic accidents each year, and tens of millions are injured. Accurate accident risk prediction has important practical significance for improving road safety, optimizing traffic management, and reducing insurance costs.

Traditional risk assessment often relies on historical accident statistics and manual experience judgment, making it difficult to capture the complex interactions of multi-dimensional environmental factors in real time. Machine learning methods can learn hidden patterns from massive historical data, providing more scientific and accurate tools for accident risk prediction.

This project uses CatBoost Regressor as the core algorithm. CatBoost is a high-performance gradient boosting decision tree library developed by Yandex, which is particularly suitable for processing structured data containing a large number of categorical features.

Dataset Features

• Road Features: Road type, number of lanes, road curvature, speed limit value
• Environmental Conditions: Lighting conditions, weather conditions, time period, whether it is a holiday, whether it is during the school term
• Traffic Information: Road sign settings, public road signs, number of historical accident reports

Target Variable

The model outputs an accident risk score, a continuous value between 0.0 and 1.0, representing the probability of a traffic accident under specific conditions.

Training Process

1. Data loading and cleaning
2. Exploratory Data Analysis (EDA)
3. Feature selection
4. Training/test set split
5. CatBoost model training
6. Model evaluation

Evaluation Results

Metric	Value	Interpretation
MAE	0.0438	The average deviation between predicted and actual values is about 4.4%
MSE	0.00317	Low error indicates good prediction stability
RMSE	0.0563	Standard deviation of error is about 5.6%
R² Score	0.8855	Explains 88.6% of the risk variation

Advantages of CatBoost

• Native support for categorical features
• Ordered Target Statistics to prevent overfitting
• Symmetric tree structure to improve training speed
• Built-in missing value handling
• GPU acceleration support

1. Intelligent Traffic Management System: Calculate road segment risk scores in real time to guide police deployment and early warning issuance
2. Insurance Actuarial Pricing: Develop refined auto insurance pricing strategies
3. Navigation Route Optimization: Recommend safer driving routes
4. Road Infrastructure Planning: Identify high-risk road segments to guide renovation

• Hyperparameter tuning (grid search/Bayesian optimization)
• Comparison with algorithms like XGBoost and LightGBM
• Build interactive web applications using Streamlit
• Introduce SHAP values to analyze feature contribution
• Connect to real-time data sources for online early warning

This project demonstrates a typical application paradigm of machine learning in the field of public safety, forming a complete technical closed loop from data integration to model deployment. CatBoost's advantages in processing structured data make it an ideal choice.

This project provides clear code structure and documentation for developers who are new to machine learning engineering, making it an excellent learning reference. With the development of the Internet of Things and intelligent transportation technology, such models will play a more important role in urban safety governance.