Cleveland Neighborhoods, Venues, and Median Home Values

Applied Data Science Capstone

Lucas Gagne
December 2019
Assigned by Coursera

Please refer to my Jupyter Notebook to see the code and dataframes mentioned.
Please refer to my Report for a downloadable pdf version of this post.

Introduction

In this project, we plan to leverage location and median home value estimate data in Cleveland, Ohio from Zillow and Foursquare API to determine any relationship between venues and the values of nearby homes.  This relationship is useful to city planners in determining zoning and its effect on home prices, as well as to commercial enterprises and the ideal locations for their businesses.

Data

First, Cleveland neighborhood data was pulled from Zillow.  The data was parsed into a dictionary from the original XML.  The dictionary was combed through to create a dataframe with the name, latitude, longitude, and 'Zindex' (median home value estimate) of each neighborhood.  Neighborhoods without 'Zindex' estimates were thrown out.



Next, venue data was pulled from Foursquare for venues within 2000 meters of each neighborhoods latitude and longitude.  This data was parsed into dictionary form from json.  This dictionary was combed through to create a dataframe with the neighborhood, name, latitude, longitude, and category for each venue.  This dataframe was converted via one-hot encoding, the Neighborhood column was added back in, and the columns reordered for ease of reading.

Methodology

The dataframes were then grouped in two ways for each neighborhood: by the frequency of each venue category, and by the sum of each venue category.  This allowed for different analysis, either grouping neighborhoods by number of venues, or by frequency of venues.  Median home value was excluded.

To aid in visualization and comparison, a 'top five' dataframe was created, with the 5 most occurring venue categories being displayed for each neighborhood, along with the estimated median home value.

First, we used K Means to cluster the neighborhoods by venue category frequency, and then sum.  We visualized the inertia of clustering for different values of k (numbers of clusters) in order to find the optimal value of k ('elbow point').  The model was fitted using the respective dataframe and a new column identifying the cluster was inserted into our 'top five' dataframe.  

A visualization was created using folium to more clearly see the geographic implications of the clustering.

Finally, a secondary approach was tested.  Using the 'one-hot' dataframe grouped by the sum of each category, an input matrix of the category counts and an output matrix of the median home value estimates were formed.  The input and output data were split into training and testing sets.  A multiple linear regression model was formed and fitted to the training data, and the r squared value was calculated for the test data. 

Results

The K means clustering grouped by sum resulted in 5 clusters.

The K means clustering grouped by mean resulted in 4 clusters.

The Multiple Linear Regression scored an average of -0.57.  This model is NOT a good fit.

Discussion

The K means clustering grouped by sum resulted in 5 clusters.  After inspecting the top 5 venue categories and median home values for each category, the following observations can be made.


The K means clustering grouped by mean resulted in 4 clusters.  After inspecting the top 5 venue categories and median home values for each category, the following observations can be made.










Based on these results, the following observations and conclusions can be made.

  1. Neighborhoods with a higher number or frequency of bars, pubs, and breweries seem to have higher home values.
  2. The zoo (and its exhibits) make for a unique neighborhood and venue distribution.
  3. Ohio City and Downtown are two unique neighborhoods with high median home value
  4. When unique neighborhoods are separated, the remaining clusters seem to stratify into a higher value market of restaurants and bars, and a lower market of fast food and discount stores. 

Conclusion

In conclusion, it is clear that there is a correlation between venues and the median home values of neighborhoods within the city of Cleveland.  This correlation is not fit to predict with linear regression, but can be understood through a K Means clustering approach.  The analysis can help entrepreneurs best place their new business, and can give cities insight into how zoning and new business are related to home values.

Comments

Post a Comment

Popular posts from this blog

Applied Numerical Methods

The following are links to the presentation and paper for ANM I The following are links to the presentation and paper  for ANM II The above two projects were completed as a graduate student taking Applied Numerical Methods.  The projects (and class) were centered around using MatLab to solve differential and integral problems.  As the only graduate student in the group, I took a leadership role in both projects, organizing and delegating as necessary.
Read more

Prime Play (In Progress)

Here is a link to my Prime Play project. Primes are relatively simple to define, and yet a formula to produce a guaranteed prime is still unknown.  I caught on to a pattern that seems to work for small numbers, but needs more testing for big numbers.  This project is a collection of Python scripts (apps essentially) that constitute different methods of finding primes.  I use the classic 'brute force' method to compare the computational efficiency of the method I'm working on.  I still need to generalize my method further before I can officially call my experiment a failure.  So far I'm not expecting a Nobel Prize. This project has made use of my knowledge of combinatorics, Python (especially classes and an iteration module), and algorithms.
Read more

Modeling the Progress and Retention of International Students Using Markov Chains

Here is a link to my Undergraduate Honors Research Project.  Below is my Abstract. Abstract International students are a small and diverse student population present in any sizable American university. One of the greatest obstacles in their path is the acquisition of the English language. English for Academic Purposes (EAP) programs, such as the English Language Institute (ELI) at the University of Akron, attempt to address this problem. By studying how this student population progresses in their academic studies, EAP programs and their associated universities can make well-informed decisions on how best to serve their English Language Learners. One way to study International students is through the use of a Markov model based on university data. The statistically calibrated model is helpful in that it can reveal at what stage of their studies students are struggling the most or succeeding the most. Our results show that ELI students are well prepared, since international stude...
Read more