An end-to-end machine-learning application for estimating California housing prices using a stacked ensemble model and an interactive Streamlit interface.
The project covers the full workflow from data preparation and model comparison to ensemble learning, evaluation, saved-model inference, and web deployment.
Try the deployed application:
https://portfolio-house-predictor.streamlit.app
Users can adjust property characteristics and receive a real-time predicted house price from the trained ensemble model.
The pipeline covers:
Data preparation → Model training → Model comparison → Stacking ensemble → Evaluation → Streamlit deployment
The goal of this project is to build a complete regression workflow rather than only train a prediction model.
The project explores how multiple regression algorithms perform on the California housing dataset and combines the strongest models through stacking to improve predictive performance.
The final trained model is then exposed through a Streamlit application so users can interact with the model directly.
This project demonstrates the complete lifecycle of a machine-learning regression application rather than only model training.
It shows how a predictive model can move through:
- data preparation;
- model comparison;
- ensemble design;
- evaluation;
- saved-model inference;
- interactive application development;
- deployment.
The practical value of the project is in connecting experimentation with a usable interface, allowing users to interact with the trained model and observe how input features affect predicted housing values.
From a portfolio perspective, it demonstrates end-to-end ML workflow design, model evaluation, deployment, and communication of results.
The final prediction system uses a stacking ensemble that combines:
- XGBoost
- LightGBM
- Gradient Boosting
A linear meta-learner combines the predictions of the base models to produce the final estimate.
This approach allows the ensemble to benefit from different model behaviors instead of relying on a single regression algorithm.
| Metric | Result |
|---|---|
| R² Score | 0.859 |
| Explained variance | 85.9% |
| RMSE | $52,800 |
| Relative R² improvement over linear-regression baseline | 49% |
The stacking ensemble achieved the strongest overall R² performance among the evaluated models.
| Model | R² Score |
|---|---|
| Stacking Ensemble | 0.8589 |
| LightGBM | 0.8559 |
| XGBoost | 0.8493 |
The reported results correspond to the evaluation configuration used in this project and should not be interpreted as universal performance guarantees.
The Streamlit interface allows users to enter housing characteristics such as:
- median income
- house age
- average rooms per household
- average bedrooms per household
- population
- average household occupancy
- latitude
- longitude
The trained model then generates an estimated property value in real time.
- End-to-end regression workflow
- Data cleaning and feature preparation
- Comparison of multiple regression models
- Gradient-boosting models
- Stacking ensemble architecture
- Saved-model inference
- Interactive property inputs
- Real-time price prediction
- Model-performance display
- Streamlit deployment
- Scikit-learn
- XGBoost
- LightGBM
- Gradient Boosting
- Python 3.11
- pandas
- NumPy
- Streamlit
- Git
- GitHub
- Python virtual environments
House_Price_Predictor/
├── Home.py
├── pages/
│ └── ...
├── Projects/
│ └── house_price_prediction/
│ ├── training and experimentation code
│ └── saved model artifacts
├── requirements.txt
├── LICENSE
└── README.md
California Housing Data
↓
Data Cleaning & Preparation
↓
Model Training
↓
Model Comparison
↓
Stacking Ensemble
↓
Model Evaluation
↓
Saved Model
↓
Streamlit Application
↓
Real-Time Prediction
MIT license