1 of 25

University of Richmond Online Data Analytics Boot Camp ��Final Project

PRESENTED BY:

  • NICK KINSLER
  • SAM LILBURN
  • MICHAEL MARONE
  • MIA MESFIN

2 of 25

Overview

  1. BACKGROUND
  2. DATA
  3. MACHINE LEARNING
  4. DATA VISUALIZATIONS
  5. RECOMMENDATIONS

3 of 25

What is a optimal market price to list?

  • Focus: Greater Richmond Area
    • Chesterfield
    • City of Richmond
    • Hanover
    • Henrico�

​

Photo Source: https://en.m.wikipedia.org/wiki/File:Richmond_%28Virginia%29.jpg

  • Listing types:
    • Single-Family
    • Condo
    • Townhouse
    • Manufactured

4 of 25

Reason why we selected topic

How things started?

  • Original topic: Can we predict the likelihood of an employee quitting?

�

Issues with original topic

  • Determined the availability of the data was difficult to find/access.
  • Would require overuse of assumptions to be made in the model.

5 of 25

Reason why we selected topic

New topic

  • What would be an optimal price to list a home for sale or bid to buy a home?�

​

Why the change?

  • No issues with availability of data.
  • Large dataset available to perform machine learning on.

6 of 25

Technology used

Python

PostgreSQL

Jupyter Notebook

Tableau

Flask

D3

7 of 25

Source of data

  • Realty Mole Property API
  • Data was pulled between November 15-17th

​

8 of 25

Description of analyst stage of project

DATA ANALYSIS PHASE

  • Ran API via Python
  • Broke out code by locality
  • Converted data into data frame
  • Exported data frames into CSV files

9 of 25

Description of analyst stage of project

DATABASE

  • Merged zip codes in Python
  • Imported into PostgreSQL via SQLAlchemy
  • Prepped data for machine learning and data visualization

10 of 25

Machine learning

MACHINE LEARNING

  • Imported CSV file from Database section
  • Ran supervised model (Linear Regression)
  • Adjustments made to improve model accuracy

11 of 25

Machine learning pipeline

1. Data

Preprocessing

2. Fit Linear Regression Model

3. Predict Property Prices

4. Model Statistics

12 of 25

Machine learning pipeline

1. Data

Preprocessing

13 of 25

Machine learning pipeline

1. Data

Preprocessing

 

14 of 25

Machine learning pipeline

2. Fit Linear Regression Model

15 of 25

Machine learning pipeline

3. Predict Property Prices

16 of 25

Machine learning pipeline

4. Model Statistics

Variable

Zip Code

Sq. Footage

Bedrooms

Bathrooms

Condo

Manufactured

Single Family

Townhouse

Coefficient

-66.02

139.64

-22117.19

39963.58

14029.78

-59941.83

24807.58

21104.47

Adjusted R2: 0.71

Y-intercept: 1585098.27

Root Mean Square Error (RMSE): $88,583.01

17 of 25

Machine learning pipeline

4. Model Statistics

Summary Statistics

Price

Predicted Price

Mean

373,272.94

374,221.51

Std. Deviation

163,416.09

137,674.18

Min

80,000.00 

(set limit)

105,928.00

Max

1,000,000.00

(set limit)

1,504,654.00

18 of 25

Machine learning pipeline

4. Model Statistics

Correlation Matrix

19 of 25

​

Result of analysis

​

DATA VISUALIZATION

​

  • Take a closer look at demographic of listings
  • Comparing the listed price to Machine Learning predictions

​

​

20 of 25

​

​

​

Demographic

Property Types

21 of 25

22 of 25

23 of 25

Result of analysis

24 of 25

Recommendations

  • County expansion
    • Further analysis to include:
      • City of Petersburg
      • Colonial Heights
      • Goochland
      • Powhatan
  • Full state integration
  • Determining optimal rental listing

25 of 25

What we could’ve done differently

  • Using radius search over Zip Code
  • Using different API
    • Zillow API

​