🏡 PropSight — Real Estate Valuation Platform

Predict residential property values with machine learning, powered by Streamlit.

---

✨ What is PropSight?

PropSight is an interactive real estate valuation web app that uses ensemble machine learning models to estimate house prices based on property features. Simply enter a home's specs — size, quality, neighborhood, year built — and PropSight returns an instant price prediction backed by three trained models: Ridge Regression, Lasso Regression, and a Gradient Boosting Regressor.

📊 Held-out test accuracy: R² = 93.68% · MAE = $12,695

---

🎯 Features

• 🔮 Instant price prediction — enter property details and get a valuation in real time
• 🤖 Three ML models — Ridge, Lasso, and Gradient Boosting with live comparison
• 📈 Key price drivers — feature importance chart pulled live from the trained GBR model
• 🏘️ Neighborhood-aware — one-hot encoded neighborhood effects trained into the model
• 📄 Technical documentation — downloadable PDF with full methodology
• 🎛️ Sensible defaults — non-user features default to dataset medians (no phantom $184M predictions)

---

🖥️ Try It Live

No installation needed — the app is deployed and ready at:

https://propsight.streamlit.app

DataSet Available at Kaggle:

Kaggle

---

🗂️ Project Structure

propsight-analysis/
├── app.py                      # 🚀 Streamlit app (entry point)
├── requirements.txt            # 📦 Python dependencies
├── notebook.ipynb              # 📓 Exploratory data analysis
├── Technical_Documentation.pdf # 📄 Full methodology writeup
│
├── data/
│   └── train.csv               # 🏠 Ames Housing dataset
│
└── models/
    ├── ridge_model.pkl         # Ridge regression
    ├── lasso_model.pkl         # Lasso regression
    ├── gbr_model.pkl           # Gradient Boosting (primary)
    ├── model_columns.pkl       # Feature schema
    ├── feature_medians.pkl     # Median defaults for non-user inputs
    ├── feature_importances.pkl # Live driver chart data
    └── metrics.pkl             # Honest held-out test metrics

---

⚡ Local Setup

# 1. Clone the repo
git clone https://github.com/Butkii025/propsight-analysis.git
cd propsight-analysis

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

# 3. Run the app
streamlit run app.py

The app will open at http://localhost:8501.

---

🧠 Models & Performance

Model	Description	Role
Gradient Boosting	500 trees, max depth 4	Primary predictor
Ridge Regression	L2-regularised linear	Comparison baseline
Lasso Regression	L1-regularised linear	Comparison baseline

All models are trained on the Ames Housing dataset and evaluated on a held-out test split.

Test R²  →  93.68%
Test MAE →  $12,695

---

🔧 Retraining the Model

If you update data/train.csv, regenerate all model artifacts with:

python train_model.py

Then push the updated models/ folder — Streamlit Cloud auto-redeploys on every push.

git add models/
git commit -m "retrain: updated model artifacts"
git push

---

🐛 Bugs Fixed vs. Original

Bug	File	Fix
Default inputs predicted $184M	app.py	Non-user features now default to dataset medians
Neighborhood had zero effect on price	train_model.py	Neighborhood one-hot columns now trained and saved
R²/MAE computed on training data	train_model.py	Reports honest held-out test metrics
"Key Drivers" chart was hardcoded	app.py	Pulled live from gbr_model.feature_importances_

---

🛠️ Tech Stack

Layer	Technology
Frontend / UI	Streamlit
ML Models	scikit-learn (Ridge, Lasso, GBR)
Data Wrangling	pandas, numpy
Visualisation	Plotly
Model Serialisation	joblib
Deployment	Streamlit Community Cloud

---

📄 License

This project is licensed under the MIT License — see the LICENSE file for details.

---

⭐ Star this repo if you found it useful! ⭐