Zillow:

---

Goals:

• Predict Property Tax assessed values ('taxcaluedollarcnt') of Single Family Properties that had a transaction during 2017.
• Improve the performance of existing model

---

Initial Thoughts

• Features related to size of the house 'sqft', 'beds', 'baths' are going to be predictive of tax value. They are going to be closely related so we may want to remove some to avoid multicolinearity.

---

Data Dictionary

Feature	Definition
beds	(float64) Number of bedrooms
baths	(float64) Number of bathrooms
sqft	(float64) finished square feet
yearbuilt	(int64) The year the house was built
poolcnt	(float64) If the property has a pool or not

Summary

• There are 56079 rows (homes)
• The target variable is tax_value, the assessed tax value of the home in dollars.

---

The Plan

Plan --> Acquire --> Prepare --> Explore --> Model --> Deliver

Acquire

* Use custom acquire module to create mySQL connection and read zillow dataset into pd.DataFrame

Prepare

* Removed columns with duplicate information
* Renamed Columns
* Handled nulls
* Changed dtypes
* Split data into train, validate and test (approx. 60/20/20)

Explore

* Vizualize data distributions of feature interaction with target
* Perform stats testing on potential features
* Choose features for the model

Model

* OLS - Multiple Regression
    * Features: 'beds','baths','sqft', 'poolcnt', 'yearbuilt'
* LassoLars
    * alpha=4
* 2nd Degree Polynomial
* GLM - Tweedie Regressor

---

Steps to Reproduce

1. Clone this repo.
2. Create env.py file with credentials to access Codeup mySQL server
3. Run notebook.

---

Next Steps:

• Zillow dataset includes many possible features. Do further feature engineering to better capture subsets of the data:
  • Distressed properties: tax_delinquency, having more than 2x bedrooms to bathrooms...
  • High-end properties: Can lot size help predict high-end properties when our features cannot?
• Spend more time investigating outliers. Eliminating the top 0.5% and bottom 0.5% had the biggest positive impact on the model. Are there erros in the data or are our features missing part of the story?
  • Low-end outliers: Are these actually SFR homes? Are there actually structures or just empty lots?