COURSERA CAPSTONE

BATTLE OF THE NEIGHBORHOODS
- SINGAPORE

Introduction / Business Problem

A group of young entrepreneurs would like to start a new cafe business in Singapore.   
However, they are unsure of where to locate their new cafe. They would their cafe to cater to a wide
group of people.  They have heard that data science might be able to give them better insights in the
location and have given me this project to use data science techniques to assist them.

Data

To approach this data science project, firstly, data regarding the neighbors in Singapore are to be retrieved
from Wikipedia (https://en.wikipedia.org/wiki/Planning_Areas_of_Singapore)

A screenshot of the relevant table is shown below:

Unfortunately, this Wikipedia page does not contain latitude and longitude information of the neighborhoods,
which will be filled in later.  

Firstly, though, the table from the Wikipedia contains information which are not required, hence columns
Malay, Chinese, Pinyin and Tamil are dropped.  A screenshot of the partial list of data at this stage is shown
below:

So back to the issue of missing latitude and longitude data, using nominatim from geopy.gelocator, the
required data is retrieved and added to each neighborhood.

During this process, three of the neighborhoods (Downtown Core, Western Islands and Museum) were not
recognized by OpenStreetMap and no latitude and longitude data were returned for these three
neighborhoods.  Upon further internet search, more detailed data regarding the neighborhood were
obtained from the following file from the Department of Statistics Singapore website:
https://www.singstat.gov.sg/-/media/files/find_data/population/statistical_tables/tablea12-2000-2018.xls

From this file, based on the largest population in the subzones of the neighborhood, three other names
were chosen (Bugis, Jurong Island and Dhoby Ghaut) to replace the three not recognized by
OpenStreetMap.

A screenshot of the partial list of neighborhood with latitude and longitude data added:

With the latitude and longitude data obtained for all neighborhoods, the latitude and longitude,
as well as the name of the neighborhood, are then used to retrieve the location data of the venues
(with a radius setting of 1000m) from Foursquare.   A total of 2991 venues in 283 unique categories
were retrieved.

Screenshot of the partial list of venues::

Screenshot of the dataset’s shape containing all the venues:

The above dataset will be used for Machine Learning of the Neighborhoods

Methodology

Firstly, the population of each neighborhood is plotted on a Folium to gain a visual representation of the
possible locations of the new cafe.  The size of the bubbles represents the size of the population of the
neighborhood. The Folium Map is shown below:

Figure 1:  Map of Neighborhood Population

Next, K-Means Clustering is used to cluster neighborhoods into five clusters based on the similarity of
venues in the neighborhoods.  The resulting neighborhoods’ cluster labels were then merged with the
neighborhoods’ latitude and longitude. A Folium map is then created to show the clusters that are similar
to each other.  The map is shown below:

Figure 2: Map of Neighborhood Clustering

In addition, as the venues returned by Foursquare consists of all categories, while this project is regarding
the location of new cafe, further data cleaning was conducted on the dataframe to remove non-food related
venues.  The venue categories were inspected in batches and non-food categories were removed.
Part of the resulting dataframe after this cleanup is shown below:

With the number of food venues found for each neighborhood, another Folium Map is created to visualize
this new information: 

Figure 3: Map of Total Food Venues in Each Neighborhood

Results

From Figure 1 Population in Singapore neighborhoods, we can see that there are neighborhoods with bigger
populations as shown by the bigger bubbles.  The top 5 neighborhoods in terms of population are: Bedok,
Jurong West,Tampines, Woodlands and Sengkang.

Table 1 below shows the Top 5 neighborhoods with the largest population

Next, from the K Means Clustering, it was found that the Top 5 most populated neighborhoods belongs to
Cluster 2, indicating that they are highly similar in terms of types of venue category.

Table 2 below shows Top 5 most populated neighborhoods in Cluster 2: 

 After obtaining the total food venues in each Neighborhood, a side-by-side comparison with the population in each Neighborhood was made.  Looking at the size of bubbles, we can tell that for some neighborhoods, the number of food venues looks small compared to the population.

Population in Each Neighborhood

Food Venues in Each Neighborhood

A calculation was then made to derive the ratio of population to food venue in each neighborhood.  

Table 3 below shows this ratio at last column.  Note that this table is sorted by population size of
neighborhood in descending order

Discussion

Through the analysis and visualization of the dataset, it was observed that there are several neighborhoods with considerably larger populations than the rest.  These neighborhoods are namely: Bedok, Jurong West, Tampines, Woodlands and Sengkang. Since the clients would like to target a wide group of customers, these five neighborhoods would have to be in their shortlist of locations.

At the same time, after clustering based on similarity of venue categories, it was found that the same five
neighborhoods also belong to the same cluster.   Thus, any of the five neighborhoods would be highly similar.
As the venues in the neighborhoods are there to provide service to the population, it can be inferred that the
population in these neighborhoods are highly similar as well.

Next, through the analysis of the ratio of population to the number of food venues (Table 3), we can see
Sengkang has highest ratio amongst the five neighborhoods.  With more people than food venues, in
comparison with other neighborhoods, it could mean that opening a cafe that would attract a large crowd. 

Conclusion

Based on the findings of the exploratory data analysis and clustering, we conclude that the top five
neighborhood that we would recommend to the clients are: Bedok, Jurong West, Tampines, Woodlands and
Sengkang.  In particular, we would also highlight that Sengkang could be the best of the five as it has the
largest population to food venues ratio.