House Price Prediction and Regression Discontinuity Analysis

This project demonstrates how to predict house prices using structured data (such as square footage, number of bedrooms, etc.) and perform Regression Discontinuity Analysis on the data. The dataset is from the King County House Sales Dataset, and the analysis focuses on exploring the impact of certain features (like house price) on house valuation.

Requirements

Make sure you have the following Python packages installed to run the code:

• pandas
• numpy
• matplotlib
• seaborn
• scikit-learn
• xgboost
• transformers (for NLP tasks)
• torch

You can install them via pip:

pip install pandas numpy matplotlib seaborn scikit-learn xgboost transformers torch

Dataset

The dataset used in this project is the King County House Sales Dataset, which contains various columns about house sales in King County. The dataset includes the following columns:

• id: Unique identifier for each house
• date: Date of the sale
• price: Price of the house
• bedrooms: Number of bedrooms
• bathrooms: Number of bathrooms
• sqft_living: Square footage of the living area
• sqft_lot: Lot size in square feet
• floors: Number of floors in the house
• waterfront: Whether the house is located on the waterfront
• view: Quality of the view
• condition: Condition of the house
• grade: Grade of the house
• sqft_above: Square footage of the house excluding the basement
• sqft_basement: Square footage of the basement
• yr_built: Year the house was built
• yr_renovated: Year the house was renovated (if applicable)
• zipcode: Zip code of the house
• lat: Latitude of the house
• long: Longitude of the house
• sqft_living15: Living area square footage of the nearest 15 neighbors
• sqft_lot15: Lot size of the nearest 15 neighbors

Steps

1. Data Preprocessing

The first step is to preprocess the data:

• Handle missing values: Missing values are imputed with the median value.
• Feature engineering: A new feature, price_per_sqft, is calculated by dividing the price of the house by the living square footage.
• Normalization and encoding: Numerical features are normalized, and categorical variables (e.g., zipcode) are one-hot encoded.

2. Predictive Model

We build predictive models to forecast house prices:

• Random Forest Regressor: A decision tree-based model that handles non-linear relationships.
• XGBoost Regressor: A gradient-boosting model known for high performance.

3. Regression Discontinuity Analysis (RDA)

A regression discontinuity design is used to analyze how house prices respond to a specific threshold:

• Threshold: The threshold is assumed to be based on house price. For example, houses priced above $500,000 are considered in a high-income group, while others are considered in a low-income group.
• Causal Inference: We create a treatment variable based on the threshold and assess how house prices behave around that threshold.

4. Evaluation

The performance of the models is evaluated using:

• RMSE (Root Mean Square Error)
• MAE (Mean Absolute Error)
• R² (Coefficient of Determination)

5. Results and Plotting

The regression discontinuity results are plotted with a scatter plot of house prices against a threshold, with a red line showing the predicted values.

How to Run

1. Download the kc_house_data.csv file and place it in the data/ directory.
2. Open the Jupyter Notebook (.ipynb file) in the code/ directory.
3. Run the notebook cells sequentially to execute the code.

Example Output:

The model will output the RMSE, MAE, and R² for both Random Forest and XGBoost models. It will also generate a plot showing the regression discontinuity analysis.

Customization

You can customize the following:

• Threshold value in Regression Discontinuity (currently set to $500,000).
• Data columns: If you have additional columns or wish to replace income with a different feature, adjust the notebook accordingly.