An interactive Streamlit web app that predicts the likelihood of road accidents based on environmental and road parameters. It uses trained machine learning models to provide mean risk predictions and a 95% confidence interval, updating live as you adjust inputs.

• 🎛️ Interactive Inputs: Adjust road and weather conditions using sliders and dropdowns.
• ⚡ Live Predictions: The app automatically recalculates the accident risk whenever inputs change.
• 📊 Confidence Intervals: Shows both lower and upper bounds of predicted risk at 95% confidence.
• 🌈 Visual Risk Gauge: Color-coded indicator (Green → Yellow → Red) based on mean predicted risk.
• 🧠 Smart Feature Engineering: Includes transformations and derived features for better model accuracy.
• 💾 Pre-trained Models: Uses three LightGBM-based regressors (model.pkl, model_lower.pkl, model_upper.pkl) and a quantile transformer for normalization.

accident-risk-predictor/
│
├── model.pkl
├── model_lower.pkl
├── model_upper.pkl
├── quantile_transformer.pkl
├── app.py
└── README.md

git clone https://github.com/tejas-130704/ML-Accident-Risk-Predication.git
cd accident-risk-predictor

python -m venv venv
venv\Scripts\activate # On Windows
source venv/bin/activate # On Mac/Linux

pip install streamlit pandas numpy matplotlib joblib scikit-learn lightgbm

Place the following files in the project directory:

• model.pkl → trained main model
• model_lower.pkl → lower bound model
• model_upper.pkl → upper bound model
• quantile_transformer.pkl → fitted QuantileTransformer

(These should match the model features used during training.)

streamlit run app.py

• model.pkl: Predicts central accident risk (mean).
• model_lower.pkl / model_upper.pkl: Estimate the lower and upper confidence bounds.
• quantile_transformer.pkl: Transforms skewed curvature feature into a normal distribution before inference.

1. Adjust inputs on the left panel (e.g., lanes, curvature, lighting, weather).

2. The right panel updates instantly showing:

   • Mean predicted accident risk
   • 95% confidence interval
   • Visual risk bar (orange band)
   • Color-coded risk label (Low, Moderate, High)

Predicted Risk (Mean): 0.627 Confidence Interval: [0.518, 0.734] Risk Level: 🔴 High Risk

This project is licensed under the MIT License — feel free to modify and use it in your own work.

• LightGBM for high-performance gradient boosting.
• Streamlit for interactive model deployment.
• scikit-learn for feature scaling and preprocessing.