Real Estate AI Suite

A full-stack ML web application for Gurgaon real estate — price prediction, analytics, and AI-powered property recommendations.

---

🔗 Links

🌐 Live App	realestate-analysis-ml.streamlit.app
🎬 Video Demo	Watch on LinkedIn

---

What is this?

An end-to-end data science project built on Gurgaon real estate data — from raw web-scraped data to a deployed multi-page Streamlit app. Covers the full ML lifecycle: data collection → preprocessing → EDA → feature engineering → model selection → deployment.

---

Architecture

Raw Data (99acres web scrape)
        │
        ▼
Data Collection Notebooks
        │
        ▼
Data Preprocessing + Outlier/Missing Value Treatment
        │
        ▼
EDA + Feature Engineering
        │
        ├──────────────────────────────────┐
        ▼                                  ▼
ML Pipeline (Price Predictor)    Similarity Matrices (Recommender)
  • Encoding + Scaling             • Cosine Similarity (3 matrices)
  • Random Forest / XGBoost        • Location distance matrix
  • Log-transform target (expm1)   • Weighted ensemble scoring
        │
        ▼
Streamlit Multi-page App (3 modules)
  ├── 📊 Analytics Dashboard
  ├── 🔮 Price Predictor
  └── 🤖 Property Recommender

---

The Three Modules

📊 Analytics Dashboard

Interactive Plotly visualizations built on cleaned Gurgaon property data.

• Sector map — price per sqft and built-up area plotted on a live Mapbox map
• Area vs Price scatter — filterable by flat/house, colored by BHK
• Price distribution — histogram + box plot comparing flats vs houses
• BHK analysis — donut chart + box plot, filtered to ≤4 BHK
• Feature word cloud — most common property amenities at a glance
• Live key stats — total properties, avg price, avg price/sqft, most common BHK

🔮 Price Predictor

Predicts Gurgaon property prices using a pre-trained scikit-learn pipeline.

Input Features (12 total):

Feature	Type
Property Type (flat/house)	Categorical
Sector	Categorical (location)
Bedrooms, Bathrooms, Balconies	Ordinal
Built-up Area (sqft)	Continuous
Servant Room, Store Room	Binary
Furnishing Type	Categorical
Luxury Category	Ordinal
Floor Category	Ordinal
Age/Possession Status	Categorical

Pipeline:

• Categorical encoding (OneHot/Ordinal via ColumnTransformer)
• Log-transform on target price → np.expm1 on output
• Output: price range = [predicted - 0.22 Cr, predicted + 0.22 Cr]
• Model stored as pipeline.pkl (loaded via Google Drive using gdown)

🤖 Property Recommender

Content-based recommender using three cosine similarity matrices.

Similarity scoring:

final_score = 0.5 × cosine_sim1  (location/proximity features)
            + 0.8 × cosine_sim2  (price features)
            + 1.0 × cosine_sim3  (amenities/property features)

• Location search: finds all properties within a user-defined radius (km) using a precomputed distance matrix
• Recommendations: top-N properties sorted by weighted similarity score
• Match scores displayed as progress bars

---

ML Details

Aspect	Details
Dataset	Gurgaon properties scraped from 99acres
Target Variable	Price (Crores INR), log-transformed
Feature Engineering	Luxury score, floor category, age buckets, built-up area imputation
Outlier Treatment	IQR-based, sector-stratified
Model Selection	Compared Linear Regression, Ridge, Lasso, Random Forest, XGBoost, Gradient Boosting
Final Model	Best performer selected via cross-validation (see Model Selection/ notebooks)
Deployment	pipeline.pkl hosted on Google Drive, loaded at runtime via gdown

---

Project Structure

Real_Estate/
├── Data Collection/              # Web scraping notebooks (99acres)
├── Data Preprocessing/           # Cleaning, merging, type fixing
├── Outlier-Missing_values/       # IQR outlier treatment, missing value imputation
├── EDA/                          # Exploratory analysis notebooks
├── Feature_selection-engineering/ # Feature creation, encoding strategy
├── Model Selection/              # Model comparison, cross-validation, final selection
├── pages/
│   ├── Analysis_App.py           # Analytics Dashboard page
│   ├── Price_Predictor.py        # Price Predictor page
│   ├── Recommender.py            # Recommender System page
│   ├── Analysis_datasets/        # CSVs and pickles for analytics
│   │   ├── data_viz1.csv
│   │   └── feature_text.pkl
│   ├── Recommender_System/       # Cosine similarity + location distance matrices
│   │   ├── cosine_sim1.pkl
│   │   ├── cosine_sim2.pkl
│   │   ├── cosine_sim3.pkl
│   │   └── location_distance.pkl
│   └── df.pkl                    # Cleaned dataframe for predictor dropdowns
├── Home.py                       # Streamlit entry point + navigation
├── requirements.txt
└── README.md

---

Setup & Run

# 1. Clone
git clone https://github.com/DEVJHAWAR11/Real_Estate.git
cd Real_Estate

# 2. Install dependencies
pip install -r requirements.txt

# 3. Run
streamlit run Home.py

Note: pipeline.pkl is downloaded automatically at runtime from Google Drive via gdown. No manual setup needed.

---

Tech Stack

Layer	Technology
Web App	Streamlit (multi-page)
ML Pipeline	Scikit-learn (ColumnTransformer + Estimator)
Data Processing	Pandas, NumPy
Visualizations	Plotly Express, Matplotlib, Seaborn, WordCloud
Model Storage	Google Drive + gdown
Deployment	Streamlit Community Cloud

---

Author

Dev Jhawar — GitHub | LinkedIn | KIIT University, CSE (CGPA: 9.66)