House Price Prediction App

This repository contains a machine learning model for predicting house prices in King County, Washington (Seattle area) based on 12 key home features. The model is built using TensorFlow/Keras and deployed via a clean Streamlit web interface.

---

Table of Contents

• Installation
• Data
• Training (HP.py)
• Prediction UI (app.py)
• Model Architecture
• Evaluation
• Deployment
• Structure
• License

---

Installation

Create a Python 3.10 virtual environment (required for TensorFlow compatibility):

python3.10 -m venv venv310
source venv310/bin/activate  # or .\venv310\Scripts\activate on Windows

Install required libraries:

pip install -r requirements.txt

requirements.txt:

streamlit
tensorflow==2.15.0
scikit-learn==1.3.2
numpy
pandas

---

Data

• Dataset: house_data.csv
• Location: King County, Washington, USA
• Features used:
  bedrooms, bathrooms, sqft_living, sqft_lot, floors, condition, grade, yr_built, zipcode, lat, long, sqft_living15
• Target: price (log-transformed during training)

---

Training (HP.py)

This script preprocesses the data, trains the neural network model using:

• Batch Size: 128
• Epochs: 300
• Dropout Rate: 0.2
• Learning Rate: 0.005

# Highlight: Feature list
features = [
    'bedrooms', 'bathrooms', 'sqft_living', 'sqft_lot', 'floors',
    'condition', 'grade', 'yr_built', 'zipcode', 'lat', 'long', 'sqft_living15'
]

• Target values are log-transformed using np.log1p(y)
• Final evaluation inverts the transform using np.expm1()
• Model and scaler are saved as HP.keras and scaler.pkl

To retrain the model locally:

python HP.py

This will regenerate HP.keras, scaler.pkl, and training_loss_plot.png.

---

Prediction UI (app.py)

• Built with Streamlit
• Users enter home features in a form
• zipcode is selected from a dropdown
• lat and long are auto-filled based on selected zipcode
• Output: estimated home price in USD

# Auto-fill lat/long based on zipcode
lat, long = zip_lat_long[zipcode]['lat'], zip_lat_long[zipcode]['long']

To launch the app locally:

streamlit run app.py

---

Model Architecture

• Input layer with 12 features
• Dense (128), ReLU
• Dropout (0.2)
• Dense (64), ReLU
• Output: Dense (1) for regression

---

Evaluation

Metric	Value
Best Cross-Validated MAE	~74,870
Final Test MAE	~89,638.35
Final Test R² Score	~0.7739

Plot: training_loss_plot.png shows training vs validation loss.

---

Structure

House-Price-App/
├── app.py                  # ✅ Main Streamlit UI
├── HP.py                   # ✅ Model training script
├── house_data.csv          # ✅ Dataset
├── HP.keras                # ✅ Trained model
├── scaler.pkl              # ✅ Scaler for user input
├── training_loss_plot.png  # ✅ Visual of training performance
├── requirements.txt        # ✅ Cleaned dependency list
├── README.md               # ✅ documentation
└── Dockerfile              # ✅ For deployment

---

For questions or contributions, contact Kyle Jackson at Kyjack66@gmail.com.