A smart AI system that predicts house prices across India. Just enter your property details — city, area, BHK, floor, furnishing — and get an instant price estimate in seconds.
This project predicts how much a house is worth based on its features like city, size, number of rooms, floor number, age, parking, etc.
It uses 5 different AI models working together like a team — each one makes its own prediction, and then a final "boss" model combines all their answers to give the most accurate price.
Think of it like asking 5 real estate experts and then taking the smartest average of their opinions.
| Metric | Value | What It Means |
|---|---|---|
| Overall Accuracy | 92.13% | Out of every ₹100, the model is off by only ₹7.87 on average |
| R² Score | 0.9682 | The model explains 96.82% of the variation in house prices |
| MAE | ₹15.2 Lakhs | Average error is about ₹15 Lakhs (on houses worth ₹20L–₹10Cr+) |
| MAPE | 7.87% | Average percentage error across all predictions |
| Model | R² Score |
|---|---|
| XGBoost | 0.9424 |
| LightGBM | 0.9705 |
| GradientBoosting | 0.9703 |
| ExtraTrees | 0.9697 |
| Deep Neural Network | 0.9611 |
| Final Stacked Model | 0.9682 |
Points close to the red dashed line = accurate predictions. Most predictions are tightly packed along this line.
Shows 4 views: scatter plot, residuals, error distribution, and per-model R² comparison.
graph TD
A[Property Details <br> City, BHK, Area, Floor, etc.] --> B[Clean & Prepare the Data]
B --> C[XGBoost <br> Prediction]
B --> D[LightGBM <br> Prediction]
B --> E[3 More Models <br> Predictions]
C --> F[Meta-Learner XGBoost <br> Combines all 5 predictions]
D --> F
E --> F
F --> G[Final Price Estimate]
classDef default fill:#1f2937,stroke:#3b82f6,stroke-width:2px,color:#f9fafb
classDef data fill:#065f46,stroke:#10b981,stroke-width:2px,color:#fff
classDef model fill:#4c1d95,stroke:#8b5cf6,stroke-width:2px,color:#fff
classDef final fill:#9f1239,stroke:#f43f5e,stroke-width:2px,color:#fff
class A,B data
class C,D,E model
class F,G final
5 Models Used:
- XGBoost — Fast and accurate tree-based model
- LightGBM — Another powerful tree model, great with large data
- GradientBoosting — Classic ensemble method
- ExtraTrees — Randomized decision trees for diversity
- Transformer DNN — Deep learning with self-attention (like how ChatGPT processes data)
All 5 predictions are combined by a Meta-Learner (stacking) for the final answer.
👉 Try the Web App on GitHub Pages
⬇️ Download the Deployed Model from Hugging Face
├── House_Price_Training.ipynb # Training notebook (run this to train)
├── House_Price_Evaluation.ipynb # Evaluation & graphs notebook
├── app.py # Flask web app (local hosting)
├── trainme.py # Training script
├── evaluate.py # Evaluation script
├── templates/
│ └── index.html # Web UI template
├── static/ # CSS & JS files
├── docs/ # GitHub Pages static site
│ └── index.html # Live demo page
├── requirements.txt # Python dependencies
├── Training_dataset/ # Training & test CSV datasets
├── performance_dashboard.png # Performance dashboard
├── performance_3d.png # 3D performance surface
├── hybrid_actual_vs_predicted.png # Actual vs predicted scatter
├── hybrid_3d_residuals.png # 3D residual analysis
└── test_predictions.csv # Predictions on test data
git clone https://github.com/AshiteshSingh/House-price-prediction.git
cd House-price-prediction
pip install -r requirements.txt
python app.pyOpen House_Price_Training.ipynb in Jupyter Notebook and run all cells.
| Technology | |
|---|---|
| 🤖 | XGBoost, LightGBM, Scikit-learn |
| 🧠 | TensorFlow / Keras (Transformer + ResNet) |
| 🔍 | Optuna (hyperparameter tuning) |
| 🌐 | Flask (web app) |
| 📊 | Matplotlib, Seaborn |
Mumbai • Delhi NCR • Bengaluru • Chennai • Hyderabad • Pune • Kolkata • Ahmedabad
Made with ❤️ by Ashitesh