IBM – Coursera

Data Science Specialization

Capstone project - Final report

Correlation between a neighborhood real estate price

and its surrounding venues

Toan Thien Le – 2018

Capstone project report 2018 Data Science Specialization - IBM

Table of content:

I. Introduction:....................................................................................................................................2
II. Data description:.............................................................................................................................3
III. Methodology:...............................................................................................................................5
1. First insight using visualization:...................................................................................................5
2. Linear Regression:........................................................................................................................6
3. Principal Component Regression (PCR):..................................................................................8
IV. Results:.........................................................................................................................................9
V. Discussion:......................................................................................................................................9
VI. Conclusion:................................................................................................................................10
References:............................................................................................................................................11
Table of Figures:...................................................................................................................................12

TOAN, LE T. 1
Capstone project report 2018 Data Science Specialization - IBM

I. Introduction:
This report is for the final course of the Data Science Specialization. A 9-
courses series created by IBM, hosted on Coursera platform. The
problem and the analysis approach are left for the learner to decide, with
a requirement of leveraging the Foursquare location data to explore or
compare neighborhoods or cities of your choice or to come up with a
problem that you can use the Foursquare location data to solve.
The main goal will be exploring the neighborhoods of New York city in
order to extract the correlation between the real estate value and its
surrounding venues.
The idea comes from the process of a normal family finding a place to
stay after moving to another city. It’s common that the owners or agents
advertise their properties are closed to some kinds of venues like
supermarkets, restaurants or coffee shops, etc.; showing the
“convenience” of the location in order to raise their house’s value.
So, can the surrounding venues affect the price of a house? If so, what
types of venues have the most affect, both positively and negatively?
The target audience for this report are:
- Potential buyers who can roughly estimate the value of a house
based on the surrounding venues and the average price.
- Real estate makers and planners who can decide what kind of
venues to put around their products to maximize selling price.
- Houses sellers who can optimize their advertisements.
- And of course, to this course’s instructors and learners who will grade
this project. Or to anyone who catch this shared on the social media
showing that I can use Python data science tools.

TOAN, LE T. 2
Capstone project report 2018 Data Science Specialization - IBM

II. Data description:

New York city neighborhoods were chosen as the observation target
due to the following reasons:
- The availability of real estate prices. Though very limited.
- The diversity of prices between neighborhoods. For example, a 2-
bedrooms condo in Central Park West, Upper West Side can cost
$4.91 million on average; while in Inwood, Upper Manhattan, just 30
minutes away, it's only $498 thousands.
- The availability of geo data which can be used to visualize the
dataset onto a map.
The type of real estate to be considered is 2-bedroom condo, which is
common for most normal nuclear families.
The dataset will be composed from the following two main sources:
- CityRealty which provides the neighborhoods average prices.
https://www.cityrealty.com/nyc/market-insight/features/get-to-
know/average-nyc-condo-prices-neighborhood-june-2018/18804
- FourSquare API which provides the surrounding venues of a given
coordinates.
The process of collecting and clean data:
- Scrap the CityRealty webpage for a list of New York city
neighborhoods and their corresponding 2-bedroom condo average
price.
- Find the geographic data of the neighborhoods. Both their center
coordinates and their border.
- For each neighborhood, pass the obtained coordinates to
FourSquare API. The “explore” endpoint will return a list of
surrounding venues in a pre-defined radius.
- Count the occurrence of each venue type in a neighborhood. Then
apply one hot encoding to turn each venue type into a column with
their occurrence as the value.
- Standardize the average price by removing the mean and scaling to
unit variance.

TOAN, LE T. 3
Capstone project report 2018 Data Science Specialization - IBM

The result dataset is a 2 dimensions data frame (Figure 1):

- Each row represents a neighborhood.
- Each column, except the last one, is the occurrence of a venue type.
The last column will be the standardized average price.

Figure 1 - Final dataset

The dataset has 50 samples and more than 300 features. The number
of features may vary for different runs due to FourSquare API may
returns different recommended venues at different points in time.
The number of features is much bigger than the number of samples.
This will cause problem for the analysis process. Detail and counter-
measurement will be discussed further in the next section.

TOAN, LE T. 4
Capstone project report 2018 Data Science Specialization - IBM

III. Methodology:
The assumption is that real estate price is dependent on the surrounding
venue. Thus, regression techniques will be used to analyze the dataset.
The regressors will be the occurrences of venue types. And the
dependent variable will be standardized average prices.
At the end, a regression model will be obtained. Along with a coefficients
list which describes how each venue type may be related to the increase
or decrease of a neighborhood’s real estate average price around the
mean.
Python data science tools will be used to help analyze the data.
Completed code can be found here: https://github.com/lethien/coursera-
ibm-ds-capstone/blob/master/Capstone_Analyze.ipynb
1. First insight using visualization:
In order to have a first insight of New York city real estate average price
between neighborhoods, there is no better way than visualization.
The medium chosen is Choropleth map, which uses differences in
shading or coloring to indicate a property’s values or quantity within
predefined areas. It is ideal for showing how differently real estate priced
between neighborhoods across the New York city map.
The map (Figure 2) shows high price in neighborhoods that located
around Central Park, Midtown and Lower Manhattan. The price reduces
further toward North Manhattan or toward Brooklyn.
Manhattan can be considered the heart of New York city. It’s where
most businesses, tourist attractions and entertainments located. So, the
venue types that can attract many people are expected to have the most
positive coefficients in the regression model.

TOAN, LE T. 5
Capstone project report 2018 Data Science Specialization - IBM

Figure 2 - New York city real estate price spread between neighborhoods
2. Linear Regression:
Linear Regression was chosen because it is a simple technique. And by
using Sklearn library, implementing the model is quick and easy. Which
is perfect to start the analyzing process.
The model will contain a list of coefficients corresponding to venue
types. R2 score (or Coefficient of determination) and Mean Squared
Error (MSE) will be used to see how well the model fit the data.
The result (Figure 3) doesn’t seem very promising. R2 score is small,
which means the model may not be suitable for the data.

TOAN, LE T. 6
Capstone project report 2018 Data Science Specialization - IBM

Figure 3 - Linear Regression result

But on the bright side, the coefficient list shows some interest and logical
information:
- “Studios” and “Eateries” both mean businesses. “Train Station”
means ease of transportation. All of which usually increase the value
of a location.
- “Bar” and “Market” sure are nice to visit sometimes but may not be a
suitable neighborhood for family with kids. “Lighthouse” and “Golf”
usually located in the rural areas. The demand for such locations is
usually low.
- “TV station”, “Cemetery”, “Laser Tag”, “Mini Golf” all give value to a
limited range of people. “Gas Station” is available everywhere. These
types of venue usually are not dicision factor when considering a
location.
Back to the model, what seems to be the problem? And what are the
possible solutions?
Looking back further to the dataset, its dimensions sizes is clearly
unbalanced, only 50 samples, and more than 300 features. Logical
steps to take are either collecting more samples or trying to reduce the
number of features.
But since there are no other public source available, increasing sample
size is not possible at the moment. So, deceasing features is the only
option for now.
And that’s why Principal Component Regression is chosen to analyze
the dataset in the next part.

TOAN, LE T. 7
Capstone project report 2018 Data Science Specialization - IBM

3. Principal Component Regression (PCR):

PCR can be explained simply as the combination of Principal
Component Analysis (PCA) with Linear Regression. [ CITATION Wik \l 1033 ]
PCR employs the power of PCA, which can convert a set of values of
possibly correlated variables into a set of values of linearly uncorrelated
variables called principal components. As the result, the number of
features is reduced while keeping most of the characteristic of the
dataset.
Then PCR use Linear Regression on the converted set to return a
coefficient list, just like in normal Regression techniques.
Again, R2 score and MSE are used to see how well the model fit the
dataset.

Figure 4 - PCR scores

The result is promising as it shows improvement over the simple Linear
Regression.
As for the coefficient list, the size has been reduced after performing
PCA. So, a dot product with eigenvectors is needed to get it back to the
original features size.

Figure 5 - Coefficient list in original size

The insight is still consistent compared to the Linear Regression’s.

TOAN, LE T. 8
Capstone project report 2018 Data Science Specialization - IBM

IV. Results:
Even though the scores seem to be improved after applying a more
sophisticate method, the model is still not suitable for the dataset. Thus,
it can’t be used to precisely predict a neighborhood average price.
Explanations for the poor model can be:
- The real estate price is hard to predict.
- The data is incomplete (small sample size, missing deciding factors).
- The machine learning techniques are chosen or applied poorly.
But again, on the bright side, the insight, gotten from observing the
analysis results, seems consistent and logical. And the insight is
business venues that can serve the needs of most normal people
usually situated in pricy neighborhoods.
V. Discussion:
The real challenge is constructing the dataset:
- Usually the needed data isn’t publicly available.
- When combining data from multiple sources, inconsistent can
happen. And lots of efforts are required to check, research and
change the data before merge.
- For data obtained through API calls, different results are returned with
different set of parameters and different point of time. Multiple trial
and error runs are required to get the optimal result.
- Even after the dataset has been constructed, lots of research and
analysis are required to decide if the data should be kept as is or be
transform by normalization or standardization.
It can be considered the most important process in the whole data
science pipeline. Which can affect the most on the result.
On the other hand, choosing the suitable technique to construct the
model is also a worthwhile process. As this report shows that, by
applying a different method, the result can be improved.

TOAN, LE T. 9
Capstone project report 2018 Data Science Specialization - IBM

VI. Conclusion:
It’s unfortunately that the analysis couldn’t produce a precise model or
showing any strong coefficient correlation for any venue type. But we
can still get some meaningful and logical insights from the result.
Doing this project helps practicing every topic in the specialization, and
thus, equipping learners with Data Science methodology and tools using
Python libraries. Also doing a real project certainly helps one learns so
much more outside the curriculum, as well as realizes what more to
research into after completing the program. And as this report shows,
there are surely a lot of things to dig into.
Some notes on the analysis result:
- This project is done by a web developer who only started self-
studying Data Science for 4 months. So please take it with a grain of
salt.
- The coefficients only show correlation, not causation. So, if your
neighborhood average price is low, please don’t go destroying the
surrounding bars and food trucks. There might be another reason.
Toward the person that went through this project, many thanks for the
time and patient.

TOAN, LE T. 10
Capstone project report 2018 Data Science Specialization - IBM

References:

Wikipedia. (n.d.). Principal component regression. Retrieved from Wikipedia:

https://en.wikipedia.org/wiki/Principal_component_regression

TOAN, LE T. 11
Capstone project report 2018 Data Science Specialization - IBM

Table of Figures:

Figure 1 - Final dataset............................................................................................................................4

Figure 2 - New York city real estate price spread between neighborhoods.....................................6
Figure 3 - Linear Regression result........................................................................................................7
Figure 4 - PCR scores..............................................................................................................................8
Figure 5 - Coefficient list in original size................................................................................................8

TOAN, LE T. 12