Predicting Tax Value

About the project

Goals

Predict the tax values of single unit properties that the tax district assessed using the property data from those whose last transaction was during the peak real estate demand months of May and August 2017.

Deliverables

Final notebook: Should contain a walkthrough of documentation and cleaned up code README: Explain what the project is, how to reproduce the work, and project planning notes. Python modules: Should contain data acquisition and preparation files.

Data Dictionary

Term	Data Type	Definition
Federal Information Processing Standard (fips)	Float64	In this data set, 6037, 6059, 6111, all are codes set by the government to recognize geographical areas.
lot_size	Float64	The square footage of the lot the property is on.
square_feet	Float64	Taken from the calculatedfinishedsqfeet column and is the total square footage of the property not to include the lot.
bedroom_count	Float64	Taken from the bedroomcnt and is the total number of bedrooms in a property. Null values were handled by filling them with the mode for the column.
bathroom_count	Float64	Taken from the bathroomcnt and is the total number of bathrooms in a property. Null values were handled by filling them with the mode for the column.
tax_amount	Float64	The amount of tax due for the given year. Nulls were handled by filling them with the mode value for the column.
tax_value	Float64	The value given by the tax assessor office to determine how much a property is worth. Nulls were handled by filling them with the mode for the column.
county	object	Using the corresponding FIPS, the name of the county was given. A county is a specific region of a state.

Hypothesis Testing

First Hypothesis 
𝐻$0$ :  Homes have the same mean tax value in each county.
𝐻𝑎 : Homes in Los Angeles have a higher mean tax value than in Ventura or Orange Counties.
alpha ( 𝛼 ): 1 - confidence level (95% confidence level -> 𝛼=.05 )
Test Used: 2 Tailed T-Test
Finding: The null hypothesis is reject meaning that homes in Los Angeles County have a higher mean value.


Second Hypothesis
𝐻0 : Number of bathrooms have no correlation with tax value. 
𝐻𝑎 : Homes with more bathrooms are correlated with higher tax values.
alpha ( 𝛼 ): 1 - confidence level (95% confidence level -> 𝛼=.05 )
Test Used: Pearson's Correlation Coefficient
Finding: Homes with more bathrooms are correlated with higher tax values.

Data Science Pipeline Used

acquire.py

1. acquire data from csv gathered from sql.

prepare.py

1. address missing data
2. address outliers
3. split into train, validate, test

explore

1. plot correlation matrix of all variables
2. test each hypothesis

feature engineering

1. split into x/y train, drop tax amount and scale the data
2. find top 3 features using KSelect Best and RFE

model

1. try different modeling algorithms: Lasso Lars, OLS, Polynomial Regression and Tweedie Regressor (GLM)
2. evaluate on train
3. select top models to evaluate on validate
4. select top model
5. run model on test to verify.

conclusion

1. summarize findings
2. make recommendations
3. next steps
4. how to run with new data.

Conclusion

• I reject the null hypothesis that homes have the same mean tax value in each county.
• I reject the null hypothesis that number of bathrooms have no correlation with tax value.
• Using Feature Engineering, the top 3 features are square_feet', 'bedroom_count', and 'bathroom_count'.
• The best performing model is the Tweedie Regressor using the top features.
• The median tax rate for Los Angeles County is 1.26, Ventura County is 1.12 and Orange County is 1.15.
• If more data and time were available, investigating number of stories, presence of hoa, and the combination of bedrooms and bathrooms as features for predicting tax value.

How to reproduce the results

You may download acquire.py and prepare.py. You will need your own env.py file with your SQL credentials in order to access the SQL server.