Car Price Prediction

• This project predicts the price of a car based on its specifications such as:

  • Company
  • Model
  • Year of Manufacture
  • Odometer Reading
  • Fuel Type
  • Transmission
  • RTO (Registration)

• A machine learning model was trained using a Pipeline that included preprocessing with One-Hot Encoding, hyperparameter tuning via GridSearchCV, and model training with CatBoost to achieve accurate predictions.

• To make the system more realistic, the predicted price is adjusted with a penalty score based on the car damage detection model (YOLO) that identifies dents, scratches, and paint damage from uploaded images.

• The car damage detection model (YOLO) generates a damage score based on the severity of detected issues, which is then used to penalize the predicted car price from the ML model, ensuring more reliable and realistic valuations.

---

Features

• Car Price Prediction using a Machine Learning Pipeline with One-Hot Encoding, GridSearchCV for hyperparameter tuning, and CatBoost for accurate results.
• Car Damage Detection using YOLO to identify Scratches, Dents, and Paint Damage from uploaded images or videos.
• Car Quality/Damage Score generation based on detected damages, used to adjust the predicted price for more realistic valuations.
• Penalty Integration: The final price is penalized according to the severity of the detected damage, ensuring fair and practical pricing.
• Streamlit UI for seamless interaction – upload images/videos, input car details, and get predictions instantly.
• Multi-Modal Input Support: Works with both Images & Videos for damage detection.
• Scalable Integration: Can be easily extended for dealership platforms, resale marketplaces, or insurance use cases.

---

Usage

1. Run Streamlit App
   • streamlit run app.py
2. Upload Car Image/Video

  Upload an image (.jpg, .png) or video (.mp4, .avi).
  The app will:
  Detect damages
  Annotate the car with bounding boxes
  Display the Car Score (/100)

3. Enter Car Details

    - Fill in the required fields:
    - Company
    - Model
    - Year of Manufacture
    - Odometer Reading
    - Fuel Type
    - Transmission
    - RTO (Region)

• Get Final Price Prediction

The ML model (CatBoost) predicts the base price.
The Car Damage Score is used to adjust/penalize the price.
The final output is a realistic car resale value displayed on the UI.

📊 Scoring Logic

• Each type of damage is given a severity weight:
• Scratch → 1
• Paint Damage → 2
• Dent → 3

The score calculation considers:

• Number of damages
• Bounding box size relative to car
• Model confidence
• Finally, the Car Quality Score is:
• Score = 100 - Normalized_Damage_Value

  So:
  100 = No Damage (Perfect Condition)
  0 = Severe Damage (Worst Condition)
  

Example Output

1. Image Example:

• Input: Car with multiple dents
• Output:
  • Annotated Image with boxes
  • Car Score: 62/100

2. Video Example:

• Real-time damage detection on uploaded car videos
• Frame-by-frame scoring

Tech Stack

• YOLOv8 (Ultralytics) – Detect car damages (scratches, dents, etc.) from images and videos
• CatBoost – Predict car price based on specs and image-derived features
• Streamlit – Web interface for uploading car specs and images, and displaying predictions
• OpenCV & Pillow – Image/video processing and annotation
• Python – Backend logic and model integration

Future Improvements

• Include insurance claim estimation based on detected damages and car value.
• Expand training dataset to include more car models, years, and diverse damage scenarios for better generalization.
• Add real-time webcam support for instant car assessment.

DEMO

• 
• video for detection
• Scoring those detection
• the complete project

Author

Sarthak Tyagi

• Machine Learning Engineer | Computer Vision Enthusiast
• LinkedIn