---

[Scenario] [Project Planning] [Key Findings] [Tested Hypotheses] [Take Aways] [Data Dictionary] [Workflow]

---

Selling homes in our new normal has just gotten easier with Zillow Offers®. Now home owners can hand over the burden of selling their property, by selling directly to us based on our state of the art Zestimate score.

The accuracy and integrity of our Zestimate score is of high importance. As a junior data scientists on the Zillow data science team, we are tasked with uncovering what drivers most affect the validity of the Zestimate score. This is measured by our target variable: logerror. Which is the difference between Zillow's estimated Zestimate and actual sale price.

logerror = log (Zestimate) − log (ActualSalePrice)

Jump to Navigation

Goal:

The goal for this project is to create a model that will accurately predict the Zestimate’s logerror. By doing so, we will uncover what features available on the Zillow Dataset are driving the amount of error.

Initial Hypotheses:

Hypothesis₁

There is a relationship between home_age and logerror.

Hypothesis₂

There is a relationship between lot_sqft and logerror.

Hypothesis₃

There is a relationship between home_value and logerror.

Hypothesis₄

County that property is in affects the mean logerror.

Project Planning Initial Thoughts:

First iteration:

An MVP; the easiest thing at each stage to move forward. Remember the MVP won't fulfill every detail of the project spec and it isn't a good use of your time to do this at first.

• Cluster: home_age, home_value
• New features:
  • home_age: current year - yearbuilt
  • tax_rate: taxamount/taxvaluedollarcnt
  • bed_bath_ratio: bedroomcnt/bathroomcnt
  • property_age_bin

Deliverables:

• Final clean, interactive Py notebook

Jump to Navigation

Exploration Key Findings:

• Ventura is a quarter of LA and OC is half of LA
• Lot size is thrown off by outliers
• home_value median price is $ 355_758
• land_value has a similar distribution to home_value, but priced lesser
• home_age is almost normally distributed.

The following features appear to have clusters to explore:

• home_age & home_value
• home_age & sqft
• lot_sqft & sqft
• home_value & sqft
• longitude & property_quality
• home_age & property_quality

Jump to Navigation

Hypothesis₁

H₀ = No correlation between home_age and logerror.

H𝛼 = There IS a correlation between home_age and logerror.

• REJECT null hypothesis.

Click to see full list.

Hypothesis₂

H₀ = No correlation between lot_sqft and logerror.

H𝛼 = There IS a correlation between lot_sqft and logerror.

• FAIL to reject null hypothesis.

Hypothesis₃

H₀ = No correlation between home_value and logerror.

H𝛼 = There IS a correlation between home_value and logerror.

• FAIL to reject null hypothesis.

Hypothesis₄

H₀ = Mean logerror is the same for small homes on small lots & Average sized homes on small lots.

H𝛼 = Mean logerror for small homes on small lots & Average sized homes on small lots are different.

• FAIL to reject null hypothesis.

Hypothesis₅

H₀ = Mean logerror is the same for properties in Los Angeles County & Orange County.

H𝛼 = Mean logerror for properties in Los Angeles County & Orange County are different.

• REJECT null hypothesis.

Hypothesis₆

H₀ = Mean logerror is the same for properties in Los Angeles County & Ventura County.

H𝛼 = Mean logerror for properties in Los Angeles County & Ventura County are different.

• FAIL to reject null hypothesis.

Hypothesis₇

H₀ = Mean logerror is the same for properties in Orange County & Ventura County.

H𝛼 = Mean logerror for properties in Orange County & Ventura County are different.

• FAIL to reject null hypothesis.

Jump to Navigation

• home_age and logerror had a weak linear relationship, at best

• lot_sqft did not have a significant effect on logerror, which we found surprising

• also surprising was the apparent lack of significance between home_value and logerror

• Out of our homemade features, small homes of all ages , large homes, and homes that are considered "best quality" seem to be drivers of logerror.

Jump to Navigation

column_name	description	key	dtype
parcelid	Unique identifier for parcels (lots)		int64
bathrooms	Number of bathrooms in home including fractional bathrooms		float64
bedrooms	Number of bedrooms in home		int64
property_quality	Overall assessment of condition of the building from best (lowest) to worst (highest)		int64
sqft	Calculated total finished living area of the home		float64
fips	Federal Information Processing Standard code - see https://en.wikipedia.org/wiki/FIPS_county_code for more details		int64
latitude	Latitude of the middle of the parcel multiplied by 10e6		float64
longitude	Longitude of the middle of the parcel multiplied by 10e6		float64
lot_sqft	Area of the lot in square feet		float64
rawcensustractandblock	Census tract and block ID combined - also contains blockgroup assignment by extension		float64
regionidcity	City in which the property is located (if any)		float64
zip_code	Zip code in which the property is located		int64
roomcnt	Total number of rooms in the principal residence		int64
unitcnt	Number of units the structure is built into (i.e. 2 = duplex, 3 = triplex, etc...)		int64
yearbuilt	The Year the principal residence was built		int64
structure_value	The assessed value of the built structure on the parcel		float64
home_value	The total tax assessed value of the parcel		float64
land_value	The assessed value of the land area of the parcel		float64
taxamount	The total property tax assessed for that assessment year		float64
logerror	The log of the difference between Zestimate value and actual sale price.		float64
transactiondate	Date property sold.		object
county	The county the property is located.		object
home_age	The current age in years of the home.		int64
logerror_quartiles	logerror distributed into 4 bins.		category

Click to see full list.

column_name	description	key	dtype
young_smhome	Indicates if the property is a young small square footage home.	1 = yes, 0 = no	uint8
middleaged_smhome	Indicates if the property is a mid-aged small square footage home.	1 = yes, 0 = no	uint8
old_smhome	Indicates if the property is an old small square footage home.	1 = yes, 0 = no	uint8
young_avghome	Indicates if the property is a young average-sized square footage home.	1 = yes, 0 = no	uint8
veteran_avghome	Indicates if the property is a mid-to-old aged average-sized square footage home.	1 = yes, 0 = no	uint8
lghome	Indicates if the property is a large-sized square footage home.	1 = yes, 0 = no	uint8
smlot_smhome	Indicates if the property is a small square footage home on a small lot.	1 = yes, 0 = no	uint8
smlot_avghome	Indicates if the property is an averaged-sized square footage home on a small lot.	1 = yes, 0 = no	uint8
smlot_lghome	Indicates if the property is a large square footage home on a small lot.	1 = yes, 0 = no	uint8
mdlot	Indicates if the property is on a medium-sized lot.	1 = yes, 0 = no	uint8
lglot	Indicates if the property is on a large-sized lot.	1 = yes, 0 = no	uint8
xllot	Indicates if the property is on an extra-large-sized lot.	1 = yes, 0 = no	uint8
structure_dollar_per_sqft	Value of the structure divided by square footage ($)		float64
land_dollar_per_sqft	Value of the land divided by square footage ($)		float64
bed_bath_ratio	Number of bedrooms divided by number of bathrooms.		float64
sqft_binned	Square footage distributed into 3 even sized bins: Small, Medium, Large.		category
LA	Indicates if property is located in Los Angeles County.	1 = yes, 0 = no	uint8
Orange	Indicates if property is located in Orange County.	1 = yes, 0 = no	uint8
avgqualityavgage	Indicates if property is a mid-aged home built of average quality.	1 = yes, 0 = no	uint8
poor_quality_old_age	Indicates if property is an old home built of poor quality.	1 = yes, 0 = no	uint8
avq_quality_young_age	Indicates if property is a young home built of average quality.	1 = yes, 0 = no	uint8
avg_quality_old_age	Indicates if property is an old home built of average quality.	1 = yes, 0 = no	uint8
bestest	Indicates if property is built of high quality.	1 = yes, 0 = no	uint8

Jump to Navigation

Please pull the repo first to use the following links to guide you through the data science pipeline. Enjoy!

1. Prep Your Repo
2. Import
3. Acquire Data
4. Clean, Prep & Split Data
5. Explore Data
   • Hypothesis Testing
6. Evaluate Data
7. Modeling
   • Identify Baseline
   • Train / Validate
   • Test

Jump to Navigation