Auto Evaluate – Used Car Price Prediction

Overview

Auto Evaluate is a machine learning project that predicts the price of used cars using Linear Regression. The model is trained on a dataset of 4009 car records sourced from Kaggle and aims to provide accurate price estimates based on various car features.

---

Objective

The goal of this project is to:

• Predict the resale price of used cars
• Understand how different features affect car pricing
• Build a complete ML pipeline from data preprocessing to deployment

---

Tech Stack

• Python
• Pandas & NumPy – Data handling
• Matplotlib & Seaborn – Data visualization
• Scikit-learn – Model building & evaluation
• Flask – Backend API
• HTML/CSS/JS – Frontend (if applicable)
• Pickle – Model serialization

---

Dataset

• Source: Kaggle

• Size: 4009 rows

• 🔗 Dataset Link: [Used Car Prices uploaded by: Taeef Najib]

• Contains features like:

  • Car brand/model
  • Year of manufacture
  • Fuel type
  • Transmission
  • Mileage
  • Engine capacity
  • Price (target variable)

---

Project Workflow

1. Data Cleaning

• Removed unwanted characters from columns (e.g., price formatting)
• Converted data types using pd.to_numeric()
• Handled missing values

2. Feature Engineering

• Applied encoding techniques:

  • One-hot encoding for low-cardinality features
  • Target encoding for high-cardinality features (like model)

• Dropped irrelevant columns

3. Data Preprocessing

• Feature scaling using StandardScaler
• Train-test split

4. Model Building

• Used Linear Regression
• Trained model on processed data

5. Model Evaluation

• Evaluated using:

  • R² Score (model performance)
  • MAE (Mean Absolute Error) (prediction error)

6. Model Deployment

• Saved model using pickle
• Built a Flask API to serve predictions
• Connected frontend with backend

---

Model Performance

• R² Score: 0.7322964983528121
• MAE: 6700.119108045312

---

---

Ui Interface

Original Price Distribution

Price Distribution after removing outliers

Model Performance

---

How to Run the Project

1. Clone the Repository

git clone https://github.com/your-username/AutoValuate.git
cd AutoValuate

2. Install Dependencies

pip install -r requirements.txt

3. Run Flask App

python app.py

4. Open in Browser

http://127.0.0.1:5000

---

Example Prediction Flow

1. User inputs car details
2. Data is preprocessed
3. Model predicts log(price)
4. Converted back using exp() transformation
5. Final price displayed

---

Challenges Faced

• Handling skewed price data
• Dealing with high cardinality features
• Avoiding negative or zero predictions
• Maintaining consistency between training and deployment pipeline

---

Future Improvements

• Use advanced models (Random Forest, XGBoost)
• Improve feature engineering
• Add more real-time data
• Deploy on cloud (AWS/Render)

---

👨‍💻 Author

Mithlesh Yeole

👨‍💻 Collaborator

Aditi Jogad

---

Acknowledgements

• Kaggle for dataset
• Scikit-learn documentation
• Open-source community

---

Note

This project is built for learning purposes and demonstrates a complete end-to-end machine learning pipeline. Flow: AUTO EVALUATE – ML PIPELINE FLOW (USED CAR PRICE PREDICTION)

1. DATA COLLECTION

• Dataset of 4009 used cars taken from Kaggle
• Contains features like model, year, fuel type, transmission, mileage, etc.

2. DATA LOADING

• Load dataset using pandas (read_csv)
• Inspect data using head(), info(), describe()

3. DATA CLEANING

• Remove unwanted characters (e.g., ₹, commas in price/mileage)
• Convert columns to numeric using pd.to_numeric()
• Handle missing values (drop or fill using median/mean)
• Fix column names (remove spaces, lowercase if needed)

4. EXPLORATORY DATA ANALYSIS (EDA)

• Check distribution of price (usually right-skewed)
• Identify outliers
• Analyze relationships between features and price
• Use plots (histogram, scatterplot, heatmap)

5. TARGET TRANSFORMATION

• Apply log transformation on price to handle skewness: y = log1p(price)
• Helps improve model performance

6. FEATURE ENGINEERING

• Handle categorical variables: a. One-Hot Encoding → for low-cardinality columns b. Target Encoding → for high-cardinality columns (e.g., model)
• Drop unnecessary columns

7. FEATURE SCALING

• Apply StandardScaler to normalize feature values
• Ensures all features are on similar scale

8. TRAIN-TEST SPLIT

• Split dataset into training and testing sets
• Typically 80% training, 20% testing

9. MODEL BUILDING

• Use Linear Regression model from sklearn
• Train model on X_train and y_train

10. MODEL PREDICTION

• Predict on X_test
• Predictions are in log scale

11. INVERSE TRANSFORMATION

• Convert predictions back to original price: price = expm1(predicted_log_value)

12. MODEL EVALUATION

• R² Score → measures goodness of fit
• MAE (Mean Absolute Error) → average prediction error

13. MODEL SERIALIZATION

• Save trained model using pickle
• Also save scaler and encoders if used

14. DEPLOYMENT (FLASK)

• Create Flask API
• Load model using pickle
• Take user input from frontend
• Apply same preprocessing steps
• Predict price and return result

15. FRONTEND INTEGRATION

• User inputs car details
• Sends request to Flask backend
• Displays predicted price

END OF PIPELINE

---