🚗 AutoVal

Used Car Resale Price Estimator

A full-stack web application that predicts the resale price of used cars in real-time.
Built with Flask, scikit-learn, and a dark-themed interactive dashboard.

---

📌 Table of Contents

• Overview
• Live Demo Preview
• Features
• Model Performance
• Tech Stack
• Project Structure
• Getting Started
• API Reference
• Dataset
• How It Works

---

🧠 Overview

AutoVal takes 10 real-world vehicle parameters and predicts the resale price using two trained ML models — a Random Forest Regressor (R² = 0.938) and a Linear Regression baseline — both trained on 15,397 real CarDekho transactions.

The app features a live metrics dashboard that shows actual vs predicted prices, residual distributions, feature importance, and prediction error by vehicle age — all fetched dynamically from the Flask backend.

---

🖥️ Live Demo Preview

The app runs locally. Below are actual screenshots of the UI:

🔷 Price Prediction Form

Real-time car price estimation with confidence interval and feature insights

📊 Metrics Dashboard

Interactive ML performance dashboard with multiple evaluation metrics and charts

---

✨ Features

Feature	Description
🎯 Real ML Prediction	Random Forest (R²=0.938) + Linear Regression trained on 15,397 records
⚡ Instant Estimates	Sub-second predictions via Flask REST API
📊 Live Metrics Dashboard	5 interactive Chart.js charts — all data fetched from backend
🔍 Price Influencers	Shows which factors pushed the price up or down
📉 Confidence Score	Per-prediction confidence interval with animated bar
🔀 Algorithm Toggle	Switch between RF and LR — charts update live
📱 Responsive UI	Works on desktop and mobile
🔁 Retrain Anytime	Run train.py with new data to update everything

---

📈 Model Performance

Both models trained on CarDekho dataset · 80/20 train-test split · random_state=42

Metric	🌲 Random Forest	📉 Linear Regression
R² Score	0.938	0.689
MAE (₹ Lakhs)	0.94	2.45
RMSE (₹ Lakhs)	1.98	4.46
MAPE	13.4%	40.1%

🏆 Top Feature Importances (Random Forest)

max_power    ████████████████████████████████████  68.1%
vehicle_age  ███████                               13.0%
mileage      ████                                   8.1%
km_driven    ███                                    5.1%
engine       ██                                     3.5%
brand        █                                      1.4%
seats        ▌                                      0.5%

💡 Key insight: Max power (bhp) is by far the strongest predictor of resale price — high-performance cars retain value significantly better than low-power equivalents regardless of age or mileage.

---

🛠️ Tech Stack

Backend

• Flask 3.0 — REST API server
• scikit-learn 1.4 — ML models
• pandas — data processing
• joblib — model serialization

Frontend

• Vanilla HTML / CSS / JS — zero frontend framework dependency
• Chart.js 4.4 — 5 interactive charts
• Google Fonts — Syne + DM Mono
• Fully dark-themed responsive UI

---

📁 Project Structure

AutoVal/
│
├── app.py                    # Flask server — API routes, loads trained models
├── train.py                  # Training pipeline — run once to generate model files
├── requirements.txt          # Python dependencies
├── cardekho_dataset.csv      # Source dataset
│
├── assets/
  ├── form.png
  └── dashboard.png
├── static/
│   └── style.css  
├── templates/
│   └── index.html            # Frontend — form + metrics dashboard
│
└── (generated after running train.py)
    ├── rf_model.pkl           # Trained Random Forest
    ├── lr_model.pkl           # Trained Linear Regression
    ├── scaler.pkl             # StandardScaler for LR
    ├── features.pkl           # Feature name list
    ├── encoder_classes.json   # Label encoder mappings
    └── metrics.json           # Evaluation metrics + chart data

---

🚀 Getting Started

Prerequisites

• Python 3.8+
• pip

1. Clone the repository

git clone https://github.com/mayankagarwal-01/AutoVal.git
cd AutoVal

2. Create and activate virtual environment

python3 -m venv venv

# macOS / Linux
source venv/bin/activate

# Windows
venv\Scripts\activate

3. Install dependencies

pip install --no-user -r requirements.txt

4. Train the models

python train.py --data cardekho_dataset.csv

Expected training output

── AutoVal Training Pipeline ──────────────────────────────────
[1/5] Loading & cleaning data...
      Loaded 15,397 rows, 11 columns after cleaning.
[2/5] Encoding categoricals...
[3/5] Splitting train / test (80 / 20)...
      Train: 12,317   Test: 3,080
[4/5] Training models...
      → Random Forest Regressor...
      → Linear Regression (with StandardScaler)...
[5/5] Saving artefacts...

── Results ────────────────────────────────────────────────────
  Random Forest:      R²=0.938  MAE=₹0.94L  RMSE=₹1.98L  MAPE=13.4%
  Linear Regression:  R²=0.689  MAE=₹2.45L  RMSE=₹4.46L  MAPE=40.1%

  Top 5 feature importances (RF):
    max_power              68.1%
    vehicle_age            13.03%
    mileage                8.08%
    km_driven              5.13%
    engine                 3.48%

  Saved: rf_model.pkl  lr_model.pkl  scaler.pkl
         features.pkl  encoder_classes.json  metrics.json

── Done. Run `python app.py` to start the server. ─────────────

5. Start the server

python app.py

6. Open in browser

http://localhost:5000

---

🔌 API Reference

Method	Endpoint	Description
GET	/	Serves the frontend
POST	/api/predict	Returns predicted price, range, confidence, factors
GET	/api/metrics?algo=rf|lr	R², MAE, RMSE, MAPE
GET	/api/metrics/scatter?algo=rf|lr	Actual vs predicted scatter data
GET	/api/metrics/residuals?algo=rf|lr	Residual histogram data
GET	/api/metrics/importance	Feature importance for both models
GET	/api/metrics/age-error	MAE by vehicle age
GET	/api/encoders	Valid values for all categorical fields
GET	/api/health	Health check

📋 POST /api/predict — Request & Response

Request body:

{
  "brand": "Hyundai",
  "vehicle_age": 6,
  "km_driven": 45000,
  "seller_type": "Individual",
  "fuel_type": "Petrol",
  "transmission_type": "Manual",
  "mileage": 18.9,
  "engine": 1197,
  "max_power": 82.0,
  "seats": 5,
  "model": "rf"
}

Response:

{
  "price": 5.42,
  "price_low": 4.99,
  "price_high": 5.85,
  "confidence": 88.3,
  "model_name": "Random Forest Regressor",
  "factors": [
    { "name": "Brand",             "impact": 1,  "label": "Hyundai" },
    { "name": "Vehicle Age",       "impact": 0,  "label": "6 yrs"   },
    { "name": "Fuel Type",         "impact": 0,  "label": "Petrol"  },
    { "name": "Transmission",      "impact": 0,  "label": "Manual"  },
    { "name": "Max Power",         "impact": 0,  "label": "82.0 bhp"},
    { "name": "Kilometres Driven", "impact": 0,  "label": "45k km"  }
  ]
}

---

📦 Dataset

CarDekho Used Car Dataset

• Source: Kaggle — nehalbirla/vehicle-dataset-from-cardekho
• Records: 15,411 → 15,397 after cleaning
• Features used: 10
• Target: selling_price (converted to ₹ Lakhs)

Feature	Type	Values / Range
brand	categorical	32 brands (Maruti, Hyundai, BMW...)
vehicle_age	integer	0 – 29 years
km_driven	integer	100 – 500,000 km
seller_type	categorical	Individual / Dealer / Trustmark Dealer
fuel_type	categorical	Petrol / Diesel / CNG / LPG / Electric
transmission_type	categorical	Manual / Automatic
mileage	float	4.0 – 33.5 kmpl
engine	integer	793 – 6,592 cc
max_power	float	38.4 – 626.0 bhp
seats	integer	2 – 9

---

⚙️ How It Works

User fills form
      │
      ▼
POST /api/predict
      │
      ▼
app.py encodes categoricals
(brand → int, fuel → int, etc.)
      │
      ▼
Builds pandas DataFrame
with correct feature names
      │
      ├──── model = "rf" ────► rf_model.predict(X)
      │                              │
      └──── model = "lr" ────►  scaler.transform(X)
                                     │
                                lr_model.predict(X_scaled)
      │
      ▼
Returns price, confidence range, influencing factors
      │
      ▼
Frontend renders result panel
Metrics charts fetch from /api/metrics/* endpoints

---

Contributors [Mayank Agarwal] [Shreyansh Verma] [Suryansh Panda] [Janhavi Maheshwari] [Yashovardhan Singh] [Khushbu Raj]