🌐 Live Demo: Click here to try the app

🏠 House Price Predictor

A production-style regression pipeline that predicts California house prices using the 1990 Census data. Covers advanced feature engineering, multiple regression models, and a beautiful interactive Streamlit web app — all containerized with Docker.

Purpose: Demonstrate regression fundamentals — feature engineering, model comparison, evaluation metrics (RMSE, MAE, R²), and building an interactive UI with Streamlit instead of a raw REST API.

---

📌 What This Project Covers

Concept	Implementation
Regression	Predicting continuous house prices
Feature Engineering	Creating 3 new meaningful features from existing ones
Model Training	Linear Regression, Random Forest, XGBoost
Evaluation Metrics	RMSE, MAE, R² Score
Model Persistence	joblib serialization
Interactive UI	Streamlit web app with sliders and inputs
Containerization	Dockerfile + docker-compose
Testing	pytest

---

🗂️ Project Structure

house-price-predictor/
├── data/
│   ├── raw/                  # Original CSV
│   └── processed/            # Cleaned data
├── notebooks/
│   └── 01_housing_eda.ipynb  # EDA and analysis
├── src/
│   ├── download_data.py      # Download California housing dataset
│   ├── preprocess.py         # Feature engineering + preprocessing
│   └── train.py              # Model training + evaluation
├── models/                   # Saved .pkl model files
├── app/
│   └── streamlit_app.py      # Streamlit web application
├── tests/
│   └── test_preprocess.py
├── Dockerfile
├── docker-compose.yml
├── requirements.txt
└── README.md

---

⚙️ Setup & Installation

1. Clone the repository

git clone https://github.com/kshah2712/house-price-predictor.git
cd house-price-predictor

2. Create virtual environment

conda create -n house-price python=3.11 -y
conda activate house-price
pip install -r requirements.txt

3. Download dataset

python src/download_data.py

4. Train the models

python src/train.py

5. Run Streamlit app locally

streamlit run app/streamlit_app.py

6. Run with Docker

docker-compose up --build

---

🎯 Dataset — California Housing

Real census data from California, 1990. Each row = one block of houses.

Feature	Description
MedInc	Median income in block
HouseAge	Median house age
AveRooms	Average rooms per house
AveBedrms	Average bedrooms per house
Population	Block population
AveOccup	Average occupants per house
Latitude	Geographic latitude
Longitude	Geographic longitude
MedHouseVal	TARGET — Median house value ($100,000s)

Engineered Features

Feature	Formula	Why
RoomsPerPerson	AveRooms / AveOccup	Measures spaciousness
BedroomRatio	AveBedrms / AveRooms	Measures layout quality
IncomePerRoom	MedInc / AveRooms	Combines income + space

---

📊 Model Results

Model	RMSE	MAE	R² Score
Linear Regression	~0.72	~0.53	~0.60
Random Forest	~0.50	~0.33	~0.80
XGBoost	~0.48	~0.31	~0.82

Lower RMSE/MAE = better. Higher R² = better (max 1.0)

---

🖥️ Streamlit App

The web app allows users to:

• Adjust house features using interactive sliders
• Get instant price predictions
• See feature importance chart
• Understand which features affect price most

---

🧪 Running Tests

pytest tests/ -v

---

🛠️ Tech Stack

• Language: Python 3.11
• ML: Scikit-learn, XGBoost
• UI: Streamlit
• Serialization: joblib
• Containerization: Docker, Docker Compose
• Testing: pytest

---

📚 Key Learnings

• Difference between regression and classification
• Feature engineering — creating meaningful features from raw data
• Regression evaluation metrics — RMSE, MAE, R²
• Building interactive ML apps with Streamlit (no frontend code!)
• Understanding feature importance in tree-based models

---

🗺️ Part of ML Learning Roadmap

This is Project 2 of 10 in a progressive ML + GenAI portfolio:

#	Project	Skills
✅ 1	Classic ML Pipeline	EDA, Sklearn, Flask, Docker
✅ 2	House Price Predictor (this project)	Regression, Feature Eng., Streamlit
3	Churn Classifier	XGBoost, SHAP, FastAPI
4	Image Classifier	PyTorch, CNN, MLflow
5	Sentiment Analyzer	HuggingFace, BERT, NLP
...	...	...

---

👤 Author

Kashyap Shah GitHub · LinkedIn