Housing Market Prediction Model

The effects of the COVID 19 pandemic have had a major effect on the housing market and home values/listing prices. Using zillow to determine average home values per month for multiple neighborhood, and combining that with Federal Reserve Interest Rates and data from the US Census, we train a machine learning model to predict future home values.

The final presentation can be viewed in the included powerpoint deck.

Team Members: Bonnie Bailey, Diane Tiblin, Joseph March

Data Sources:

• Zillow Home Value Index - Metro US: https://www.zillow.com/research/data/
• FRED Economic Data: https://fred.stlouisfed.org/series/FEDFUNDS
• US Census: https://www.census.gov/data/developers/data-sets.html

Resources:

• AWS S3 Buckets
• Census and json
• Google Colab
• Scikit-learn
• PySpark
• Pandas

Coding:

• Pulled API data from US Census.
• Cleaned the data by modifying headers, formatting data and converting to rates.
• Uploaded Zillow and Federal Funding data to AWS in S3 buckets.
• Pulled Zillow and Federal Funding data using PySpark.
• Cleaned MSA columns so Census data would match Zillow in order to join data frames together.
• Converted Federal Funding data into lists to join to the merged data frame with date columns.
• Removed duplicate and non-zero values from the merged data frame. Performed one-hot encoding on the categorical data.
• Scaled the data.
• Created and tested linear regression model.

Model Performance

We tested several versions of the model before coming to our final version.

Challenges

• Difficulty finding free data on property sales.
• Slight narrowing of our data source with the elimination of fields with null data or duplications.
• Data was limited to metropolitan areas and the suburbs surrounding.

Results/Findings

A model can be created to predict the market value to within 99.99% accuracy within a metropolitan area.

Note: The housing market is naturally unpredictable as a result of human behavior.

How Could We Expand The Project?

• Currently the model predicts one month ahead. With additional work the model could be expanded to predict further into the future.
• Currently the model predicts averages within a metropolitan area. With funding, data could be pulled for individual properties.

Contacts

• Bonnie Bailey: [email protected]
• Diane Tiblin: [email protected]
• Joseph March: [email protected]