- Justin Evans
- Yvette Ibarra
What is driving the errors in the Zestimates?
This team has been tasked to collect, clean and alayze Zillow data from 2017 in order to improve a previous prediction model that was designed to predict the Home Sale Value for Single Family Properties based on available realestate data.
- Use clusters to assist in our exploration, understanding, and modeling of the Zillow data, with a target variable of logerror for this regression project.
- Discover key attributes that drive error in Zestimate logerror.
- Use those attributes to develop a machine learning model to predict impact on logerror.
- Can be accomplished using a local env.py containing user, password, host information for access to the Codeup SQL database server.
- All other step by step instructions can be found by reading and running the below Jupyter Notebook file located in our Codeup-Justin-Evans-Yvette-Ibarra github repository found here: (https://github.com/Codeup-Justin-Evans-Yvette-Ibarra/project_zillow_team).
- Final_Report_Zillow_Team_Project.ipynb
- wrangle.py
- explore.py
- model.py
Our initial thoughts are that outliers, age, and location are drivers of errors in Zestimate.
-
Acquire data from Codeup database
-
Prepare data
-
Explore data in search of drivers of logerror
-
Answer the following initial question:
-
Question 1. Do Longitude and Lattitude have a relationship with eachother and our target feature of Zillow Zestimate logerror?
-
Question 2. Is there a relationship between our loc_clusters feature, and each of the five independent clusters 0-4 as binary categorical features, with logerror?
-
Question 3. Is there a relationship between log_error and tax delinquency?
-
Question 4. Do homes that are younger than 81 years have more log error?
-
Question 5. Do the clusters have a relationship with logerror and squarefeet of the home?
-
-
Develop a Model to predict error in zestimate.
- Use drivers identified in explore to build predictive models of error using...
- Evaluate models on train and validate data using RMSE (Root mean square Error)
- Select the best model based on the least RMSE
- Evaluate the best model on test data
-
Draw conclusions
| Feature | Description |
|---|---|
| parcelid | Unique Property Index |
| bedrooms | Number of bedrooms in home |
| bathrooms | Number of bathrooms in home including fractional bathrooms |
| calculatedbathnbr | Continuous float64 count of Bathrooms (including half and 3/4 baths) |
| fullbathcnt | Count of only Full Bathrooms (no half or 3/4 baths) |
| age | The age of the home in 2017 |
| yearbuilt | The Year the principal residence was built |
| Feature | Description |
|---|---|
| has_basement | Basement on property (if any = 1) |
| has_deck | Deck on property (if any = 1) |
| has_fireplace | Fireplace on property (if any = 1) |
| has_garage | Garage on property (if any = 1) |
| has_hottuborspa | Hot Tub or Spa on property (if any = 1) |
| has_pool | Pool on property (if any = 1) |
| optional_features | Property has at least one optional feature listed above (if any = 1) |
| has_tax_delinquency | Property has had Tax Delinquncy (if any = 1) |
| Feature | Description |
|---|---|
| fips | Federal Information Processing Standards (FIPS), now known as Federal Information Processing Series, are numeric codes assigned by the National Institute of Standards and Technology (NIST). Typically, FIPS codes deal with US states and counties. US states are identified by a 2-digit number, while US counties are identified by a 3-digit number. For example, a FIPS code of 06111, represents California -06 and Ventura County -111. |
| state | This is the two letter abbreviation for the State as defined by the FIPS code |
| county | FIPS code for california counties |
| la_county | fips: 6037; Categorical Binary Feature for Los Angeles County (if True = 1) |
| orange_county | fips: 6059; Categorical Binary Feature for Orange County (if True = 1) |
| ventura_county | fips: 6111; Categorical Binary Feature for Los Angeles County (if True = 1) |
| longitude | Longitude is a measurement of location east or west of the prime meridian at Greenwich, London, England, the specially designated imaginary north-south line that passes through both geographic poles and Greenwich. Longitude is measured 180° both east and west of the prime meridian. |
| latitude | Latitude is a measurement on a globe or map of location north or south of the Equator. Technically, there are different kinds of latitude, which are geocentric, astronomical, and geographic (or geodetic), but there are only minor differences between them. |
| zipcode | A group of five or nine numbers that are added to a postal address to assist the sorting of mail. |
| regionidcounty | Location code that identifies the Region and County of the property within the state |
| rawcensustractandblock | Raw unformatted data that identifies Census tracts and blocks |
| censustractandblock | Census tracts are small, relatively permanent geographic entities within counties and Block numbering areas (BNAs) are geographic entities similar to census tracts, and delineated in counties (or the statistical equivalents of counties) without census tracts. |
| Feature | Description |
|---|---|
| sqft | Calculated total finished living area of the home |
| lotsizesquarefeet | Calculated area of land lot belonging to parcel |
| Feature | Description |
|---|---|
| tax_value_bldg | The total tax assessed value of the structure |
| tax_value | The total tax assessed value of the parcel |
| tax_value_land | The total tax assessed value of the land |
| taxamount | The total tax fee to be collected on the parcel |
| Feature | Description |
|---|---|
| log_error | This is the logerror of the Zillow Zestimate |
| Feature | Description |
|---|---|
| loc_clusters | Created using 'longitude', 'latitude', 'age' with n_clusters = 4 |
| cluster_price_size | Created using 'taxamount', 'sqft', 'lot_sqft' with n_clusters = 4 |
| cluster_delinquency_value | Created using ‘tax_value’, ‘sqft’,‘has_taxdelinquency’ with n_clusters = 4 |
- Longitude latitude and age seem to be drivers of log_error.
- Under Cluster: Location sub-clusters 2, 3, and 4 are not driving as much error as sub-cluster 0 and sub-cluster1
- Properties that have tax delinquency on average have a higher log error.
- Under Cluster:Price & Size there is a grouping with a large spread of log error in sub-cluster 0 and sub-cluster 1
The final model has an RMSE of 0.182215 and did not beat baseline at 0.171387.
Although many of our selected features showed a statistically significant relationship with our target variable; our models indicated that none of the features were as impactful as we had initially believed.
- Our current best model did not beat baseline and is not ready for implementation.
- We would like more time to further explore our current data and possibly acquire additional data. This would allow us to possibly improve our current model.