🚗 Used Car Price Prediction – End-to-End ML & Flask App

📌 Project Overview

This project is an end-to-end machine learning system for predicting used car prices. It follows a production-oriented workflow starting from raw data preprocessing and feature engineering, through model training and evaluation, and finally preparing the system for deployment as a Flask-based web application.

---

🎯 Objectives

• Clean and preprocess real-world used car data
• Perform feature engineering using meaningful raw features
• Train and evaluate multiple regression models
• Select the best-performing model based on objective evaluation metrics
• Prepare the project for full-stack deployment using Flask

---

📂 Dataset

• Source: Kaggle – Used Cars Dataset
• Kaggle dataset Link : https://www.kaggle.com/datasets/austinreese/craigslist-carstrucks-data/
• Target Variable: price
• The dataset contains vehicle specifications, condition, usage history, and location data
• download dataset (vechiles.csv)file and put it in '../data/raw' and start by running notebooks one by one

---

🛠️ Tech Stack

• Language: Python
• Libraries:
  • pandas, numpy
  • scikit-learn
  • xgboost
  • matplotlib, seaborn
  • tqdm
• Deployment: Flask (planned)

---

⚙️ Project Workflow

1. Exploratory Data Analysis (EDA)
2. Data Cleaning
   • Handling missing values
   • Removing outliers
3. Feature Engineering
   • Creating derived features (e.g. car_age)
   • Preparing categorical features for encoding
4. Model Training & Evaluation
5. Model Selection
6. Model Saving for Deployment
7. Full-Stack Application (Flask)

---

🧠 Features Used (Before Encoding)

🎯 Target Variable

• price – vehicle selling price

---

🔢 Numerical Features

• year – manufacturing year
• odometer – mileage
• cylinders – number of engine cylinders

---

🚘 Categorical Features

• manufacturer – car brand
• model – vehicle model
• condition – overall vehicle condition
• fuel – fuel type
• transmission – transmission type
• drive – drivetrain
• type – vehicle body type
• size – vehicle size category
• paint_color – exterior color
• title_status – legal title status
• state – vehicle location (US state)

Categorical features are encoded using one-hot encoding during the preprocessing stage inside the machine learning pipeline.

---

🤖 Models Trained

The following regression models were trained and evaluated:

• Linear Regression
• Ridge Regression
• Lasso Regression
• ElasticNet
• K-Nearest Neighbors Regressor
• Decision Tree Regressor
• Random Forest Regressor
• Gradient Boosting Regressor
• XGBoost Regressor

Note: Support Vector Regression (SVR) was intentionally excluded due to scalability limitations and poor suitability for deployment.

---

📊 Model Evaluation Metrics

Models were evaluated using:

• MAE (Mean Absolute Error) ↓
• RMSE (Root Mean Squared Error) ↓
• R² Score ↑

---

🏆 Model Selection

After comparing all trained models, Random Forest Regressor was selected as the final model.

Reasons for Selection:

• Lowest MAE (average pricing error)
• Lowest RMSE (penalizing large errors)
• Highest R² score (~0.89)
• Strong generalization performance

The Random Forest model provided the best balance between accuracy, robustness, and production readiness.

---

📈 Final Model Performance

Metric	Value
MAE	~2200
RMSE	~4700
R² Score	~0.89

---

💾 Model Saving

The final trained model was saved using joblib for deployment:

joblib.dump(model, "models/used_car_price_model.pkl")


## Deployment Strategy:

- User inputs raw feature values through a web form

- Backend handles preprocessing and encoding
 
- Model returns real-time price prediction
 
- This design ensures clean user input and consistent preprocessing during inference.