Project goals

• Find the key drivers of property value for single family properties.

Project description

• I've been asked to come up with a new model for determining tax value using features created by me for single family properties. This use a dataframe from Zillow in 2017.

The Plan

Initial hypotheses

• Split into different counties since location could be a big factor in home value
• Finished sqft and amount of bedrooms will make a difference in price
• Look into bathrooms as well

Acquire:

• Acquire the data from SQL
• You will need to use the properties_2017, predictions_2017, and propertylandusetype tables.

Explore:

• What affect does amount of bedrooms have on home values?
• Are the amount of bathrooms dependent of home value or independent?
• How much does finished area affect home values?
• Does having a garage change the price?

Modeling:

• Use drivers in explore to build predictive models of different types
• Evaluate models on train and validate data
• Select the best model based on accuracy
• Evaluate the test data

Data dictionary:

Feature	Definition
Bedrooms	Number of Bedrooms
Bathrooms	Number of Bathrooms
Finished Area	Usable Square Footage of Home
Home Value	Price of Home
Year Built	Year house was built
Lot Area	Square Footage of the entire property lot
County	County location of home
Garage Fits	Number of Cars that Fit in Garage
Garage Area	Square Footage of the Garage
Transaction Date	Last time the house was sold in 2017

How to Reproduce

• Clone this repo
• Acquire data from MySql (Should make a zillow.csv after)
• Run Notebook

Key findings

• Bedrooms, Bathrooms, and Finished_area were the best features to use. Garages did not make the cut due to too much missing data and too many assumptions.
• Model 4, Polynomial Regression with a degree of 3 worked best.

Takeaways and Conclusions

• After running scaled data through the model
• Test data ran on Model 4
  • 39.7% accuracy overall
• Still quite low and only within 338k of the correct home value price which is alot of error.

Recommendations

• Recommend splitting data into counties and adding more features to the data collected in order to potentially predict Home Values in the future.
• Simply having just bedrooms, bathrooms, and finished area are not enough to predict home values.