🚦 Traffic Accident Prediction
This project develops a predictive model to estimate daily traffic accident counts in Toronto based on weather and geographic data. It integrates traffic collision data from the Toronto Police Service with weather data from Environment Canada.
Tools & Techniques:
- Data preprocessing with Power Query and Pandas
- Exploratory Data Analysis (EDA)
- Correlation and regression analysis
- Machine Learning: Random Forest, Gradient Boosting, Linear Regression
- Model evaluation using RMSE and R² metrics
Key Insights:
- Weather variables alone are weak predictors; more features (e.g., traffic volume) are needed
- Extreme accident days are outliers, challenging for models to predict
- Random Forest outperformed other models in RMSE and generalizability
Business Impact: The project provides valuable tools for urban planning, emergency services, and risk mitigation based on weather conditions. The application forecasts accident risk in specific locations given current weather data.
Interactive App:
Visit App: predictive-accidents-toronto.vercel.app
📊 Notebooks: