Project Zillow

---

Use an accurate machine learning model to forecast the property tax assessment value for Zillow's single-family residential properties with transaction dates in 2017.

Project Description

As the foremost real estate platform in the United States, Zillow and its affiliated entities offer users a seamless, transparent, and highly accessed platform for selling, purchasing, renting, and securing financing. I have decided to delve into the multiple factors that impact the evaluation of property tax assessments.

Project Goal

• Recognize the elements that exert influence on property values in a specific market.
• Employ these elements to construct a machine learning model capable of delivering accurate property value forecasts.
• This information will bolster our understanding of the assessment procedure for single-family property taxes and their appraisals.

Initial Thoughts

I propose that elements like room count, square footage, and location are likely to emerge as significant factors affecting tax-assessed property values.

The Process

1. Acquire data from Codeup MySQL DB

2. Prepare data

• Create developed columns/features from existing data

3. Explore data in search of drivers of property value

• Answer the following initial questions

  • Is there a correlation between area and prop_value?
  • Is there a correlation between age and prop_value?
  • Is there a correlation between the room count and prop_value?
  • Is there a difference in average prop_value between counties?

• Provide responses to the initial inquiries:

• Does a relationship exist between the property's size and its value?

• Is there a connection between the property's age and its value?

• Does the number of rooms in a property correlate with its value?

• Are there variations in average property values among different counties?

4. Develop a Model to predict property value

• Utilize the identified drivers from the exploration phase to construct various types of predictive models.
• Assess the models using training and validation data.
• Choose the most suitable model by considering its Root Mean Square Error (RMSE) and R-squared ($R^2$) performance metrics.
• Evaluate the selected model's performance on the test dataset.

5. Draw conclusions

Data Dictionary

DF Name	DB Name	dtype	Definition
age	NA (2017 - yearbuilt)	int	Age of the property
beds	bedroomcnt	int	Total bedrooms
baths	bathroomcnt	int	Total bathrooms
area	calculatedfinishedsquarefeet	int	Square footage of building
prop_value	taxvaluedollarcnt	int	Tax value of property (USD)
county	fips	str	County in California
latitude	latitude	float	Latitude Coordinate
longitude	longitude	float	Longitude Coordinate

FIPS County Codes:

• 06037 = LA County, CA
• 06059 = Orange County, CA
• 06111 = Ventura County, CA

Steps to Reproduce

1. Clone the repo
2. If you have access to Codeup's MySQL DB:
   • Save env.py in the repo with user, password, and host variables
   • Run notebook
3. If you don't have access:
   • Request access from Codeup
   • Do step 2

Conclusions

Takeaways and Key Findings

• Property value doesn't necessarily correlate with county, but more so the locations that are in each county
• Denser population and living area typically leads to higher property values.
• The property's location plays a significant role in determining its value.
• There is room for further enhancement and feature selection in the model's performance.

Recommendations and Next Steps

• More time is needed to work on features to better improve the model
  • latitude and longitude could hopefully give insights into cities and neighborhoods with higher or lower property values
  • potentially better data collection for some of the features