Zillow Regression Project

Building a Predictive Model for Property Values

About the Project

Big Idea: Can I build a machine learning model using regression to predict a single unit property's value that performs better than the baseline model?

How is success defined: The primary goal of this project is to utilize and hone the tools I have acquired as a data scientist by working through the data science pipeline. The secondary goal is to develop a model utilizing statistically justifiable features that performs at least as well if not better than the baseline model for predicting a single unit property's value.

Project Presentation in Tableau: Click here to view my project slide deck I created and presented using Tableau.

What is included in this repo?

• This Readme file:

  • Project Overview and Key Takeaways
  • Data Dictionary
  • Skills and tools necessary for replication
  • Outline of process and tips for replication

• Necessary Modules as .py files -acquire.py -prepare.py -evaluate.py -explore.py -preprocess.py

• Report notebook with highlights from my process

  • Key takwaways from Data Acquisition
  • How I cleaned and prepared the data
  • Exploratory analysis and key questions
  • Hypotheses regarding drivers of the target and statistical testing
  • Visualizations of features and targets (univariate, bivariate and multivariate)
  • Feature selection and feature engineering
  • Development of a baseline model
  • Model development and evaluation metrics

• XLSL file for exporting dataframe to Tableau for visulaizations

Data Dictionary

Target	Description	Data Type
'taxvaluedollarrcnt'	The total tax assessed value of the parcel (THE VALUE)	float64

Features	Description	Data Type
'parcelid'	Index: Unique identifier for each property	int64
'bathroomcnt'	Indicates the number of bathrooms a property has and includes fractional bathrooms	float64
'bedroomcnt'	Indicates the number of bedrooms a property has	float64
'calculatedbathnbr'	Indicates the number of bathrooms and includes fractional bathrooms	float64
'calculatedfinishedsquarefeet'	Calculated total finished living area of the property	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	float64
'longitude'	Longitude of the middle of the parcel	float64
'tax_rate'	Calculated tax rate for the property	float64
'bath_per_sqft'	Calculated baths per finished square feet of the living area	float64
'structuretaxvaluedollarcnt'	tax value of the finished living area on the property	int64
'landtaxvaluedollarcnt'	Tax value of the land area of the parcel	int 64
'taxamount'	The total property tax assessed for that assessment year	int64
'propertylandusetypeid'	Type of land use the property is zoned for	int64
'propertylandusedesc'	Description of the allowed land uses (zoning) for that property	object

Project Replication Tips

Technical Skills

• Python

  • Pandas
  • Seaborn
  • Matplotlib
  • Numpy
  • Sklearn -Preprocessing -Feature_Selection -SelectKBest -Recursive Feature Elimination

• SQL

• Statistical Analysis

  • Descriptive Stats
  • Hypothesis Testing
    • Pearsons Correlation Testing

• Regression Modeling

  • Linear Regression
  • LASSO + LARS
  • Generalized Linear Model (TweedieRegressor)
  • Baseline Accuracy

Modules Included in Repo

• acquire.py
• prepare.py
• explore.py

Project Process

Trello Board My process and steps that I took can be viewed on my Trello board at this link: https://trello.com/invite/b/5UAXYMrN/6df7dfe75d97cd4c5f9be7738376d19f/zillow-regression-project

Key Findings

• Number of Bathrooms, Number of Bedrooms, and Finished Living Area Square Feet are the top three drivers of property value.
• The mean baseline is $449682.262 and the baseline model performance can be evaluated by an RMSE of $348,740
• The GLM model performs better than the baseline model with a RMSE of $289,445 on out of sample data.