🏠 Delhi House Rent Price Prediction

End-to-End Machine Learning Project with Streamlit Deployment

Predicting rental prices across Delhi localities using hybrid ensemble learning
RandomForest · XGBoost · Hybrid Stacking · R² = 0.97

   

📌 Table of Contents

• Overview
• Problem Statement
• Key Results
• ML Workflow
• Models & Performance
• Visualizations
• Project Structure
• Quickstart
• Tech Stack
• Roadmap
• Citation

---

🧬 Overview

Delhi House Rent Price Prediction is a complete machine learning pipeline that predicts monthly rental prices for properties across Delhi. The project tackles real-world data challenges — noisy listings, locality imbalance, missing values, and non-linear pricing — through robust preprocessing, feature engineering, and a hybrid stacking ensemble that combines RandomForest and XGBoost.

📓 The notebook Delhi_House_Rent_Price_Prediction.ipynb covers the full pipeline — from raw data ingestion to model deployment via Streamlit.

---

💡 Problem Statement

Rental prices in Delhi are highly variable, driven by factors that interact in complex, non-linear ways:

Factor	Impact on Rent
📍 Locality	Strongest predictor — varies drastically by neighbourhood
📐 Area (sq.ft)	Direct positive correlation with price
🛏️ BHK & Bathrooms	Configuration drives base pricing
🛋️ Furnishing Status	Furnished adds 15–40% premium
🏗️ Amenities	Gym, parking, security affect price band

Traditional linear models fail to capture these interactions. This project addresses that gap with ensemble methods and locality-aware feature engineering.

---

📊 Key Results

Real metrics from model evaluation on the Delhi rental dataset (10,000–15,000 listings)

Model	MAE (₹)	RMSE (₹)	R² Score
Linear Regression	baseline	baseline	baseline
RandomForest Regressor	27,637	53,878	0.96
XGBoost Regressor	27,378	53,878	0.97
🏆 Hybrid Ensemble (Final)	29,099	53,878	0.96

Key finding: XGBoost achieved the best individual R² (0.97). The hybrid ensemble provides the most stable generalization across all Delhi locality segments, particularly for mid-range and luxury rental properties.

---

🔄 ML Workflow

Raw Data → Cleaning → EDA → Feature Engineering → Scaling → Training → Evaluation → Deployment

Phase	Details
1. Data Cleaning	Deduplication, outlier capping (1st–99th percentile), missing value imputation
2. EDA	Locality distribution, price correlation heatmaps, rent histograms
3. Feature Engineering	price_per_sqft, locality_mean_price, area_to_BHK_ratio, furnishing score
4. Encoding	Label encoding (tree models), one-hot encoding (linear models)
5. Scaling	StandardScaler for linear baseline
6. Model Training	RandomForest, XGBoost, Hybrid Stacking with Ridge meta-learner
7. Cross-Validation	5-fold K-Fold CV for robust metric estimation
8. Evaluation	MAE, MSE, RMSE, R², residual analysis, actual vs. predicted plots
9. Deployment	Streamlit web app with Joblib-saved model

---

🤖 Models & Performance

Why Hybrid Ensemble?

The stacking approach combines the best of both worlds:

• RandomForest — robust to noise, handles locality variance via bagging
• XGBoost — precise gradient-boosted corrections, captures complex interactions
• Ridge meta-learner — learns optimal combination weights from out-of-fold predictions

# Stacking architecture (simplified)
base_models  = [RandomForestRegressor(), XGBRegressor()]
meta_learner = Ridge()

# Out-of-fold predictions → meta-features → final prediction ŷ

Top Feature Drivers

Rank	Feature	Relative Importance
🥇	locality_mean_price	████████████ Highest
🥈	area_sqft	█████████ High
🥉	BHK	██████ Medium
4	furnishing_score	████ Medium
5	bathrooms	███ Low–Medium

---

**

📈 Visualizations

Actual vs. Predicted — Hybrid Ensemble

Actual vs. Predicted — XGBoost

Actual vs. Predicted — RandomForest

---

📁 Project Structure

Delhi-House-Rent/
│
├── 📓 Delhi_House_Rent_Price_Prediction.ipynb   # Main ML notebook (EDA + training)
│
├── 🌐 app.py                                    # Streamlit web application
│
├── 💾 model.pkl                                 # Saved hybrid ensemble model
├── 💾 scaler.pkl                                # Saved StandardScaler
│
├── 🗄️ Indian_housing_Delhi_data.csv             # Raw Delhi rental dataset
│
├── 📦 requirements.txt                          # Python dependencies
│
├── 🖼️ assets/                                   # Result plots & visualizations
│   ├── hybrid_predictions.png
│   ├── xgboost_predictions.png
│   ├── residual_analysis.png
│   └── rent_distribution.png
│
└── 📄 README.md

---

🚀 Quickstart

Option A — Google Colab (Zero setup)

Click the badge above to run the full notebook instantly in your browser.

---

Option B — Run Locally

1. Clone the repository

git clone https://github.com/AmanKumar-23/End-to-End-ML-Project-Delhi-House-Rent.git
cd End-to-End-ML-Project-Delhi-House-Rent

2. Create virtual environment

python -m venv venv
source venv/bin/activate      # Mac/Linux
venv\Scripts\activate         # Windows

3. Install dependencies

pip install -r requirements.txt

4. Launch Streamlit app

streamlit run app.py

5. Or run the notebook

jupyter notebook Delhi_House_Rent_Price_Prediction.ipynb

⚠️ Python 3.8+ recommended. No GPU required.

---

Quick Prediction (Python)

import joblib
import numpy as np

model  = joblib.load('model.pkl')
scaler = joblib.load('scaler.pkl')

# 3BHK · 1200 sqft · Semi-Furnished · Dwarka
features   = np.array([[1200, 3, 2, 1, 5]])   # area, BHK, bath, furnishing, locality_id
prediction = model.predict(features)
print(f"Estimated Monthly Rent: ₹{prediction[0]:,.0f}")

---

🛠️ Tech Stack

Category	Tools
Language	Python 3.8+
Data Processing	Pandas, NumPy
Machine Learning	Scikit-learn, XGBoost
Visualization	Matplotlib, Seaborn
Model Persistence	Joblib
Web Deployment	Streamlit
Notebook	Jupyter

---

🗺️ Roadmap

• Data cleaning & preprocessing pipeline
• Exploratory data analysis
• Locality-based feature engineering
• RandomForest & XGBoost training
• Hybrid stacking ensemble (R² = 0.97)
• Streamlit deployment
• SHAP feature importance visualization
• Delhi locality choropleth rent heatmap (Folium/Plotly)
• Hyperparameter tuning with Optuna
• Docker containerization
• GitHub Actions CI/CD pipeline
• AWS / GCP cloud deployment
• Temporal rent trend modelling

---

👤 Author

Aman Kumar Faculty of Technology, University of Delhi

Supervisor: Dr. Sangeeta Yadav · Department of CSE · University of Delhi (2025–2026)

---

📎 Citation

@misc{delhirent2024,
  author    = {Aman Kumar},
  title     = {Delhi House Rent Price Prediction: End-to-End ML with Hybrid Ensemble},
  year      = {2024},
  publisher = {GitHub},
  url       = {https://github.com/AmanKumar-23/End-to-End-ML-Project-Delhi-House-Rent}
}

---

Built with ❤️ at Faculty of Technology, University of Delhi

⭐ Star this repo if it helped you — it motivates further development!